Stochastic Systems AAU · YouTube · 10:14
This 120-second introductory clip presents stochastic processes, also called random processes, as probability models for signals or functions in time or space. The lecturer first establishes the terminology, then writes an informal definition and lists four motivating examples: speech signals, received power in radio transmission, accelerometer data, and noise signals. The remainder of the clip develops the speech-signal example in detail by showing two printed amplitude-versus-sample-index plots of the phrase "Stochastic Processes". The lecturer argues that such a real-world signal is not naturally described by a simple polynomial or cosine formula, and that two recordings of the same spoken phrase are not exactly identical even though they sound similar. This variability across repeated realizations is used to motivate modeling signals probabilistically rather than as one fixed deterministic function. This 120-second introductory segment uses two printed waveform plots of the phrase "Stochastic Processes" to motivate stochastic modeling: the recordings sound similar but differ in detail. The presenter then lists three engineering applications—compression, denoising, and speaker recognition—each relying on distinguishing relevant structure from variation. At 01:31 the example changes to telecommunications, showing a plot of normalized received power versus traveled distance from a mobile radio measurement, again emphasizing fluctuating signals that call for probabilistic description. This 120-second lecture fragment introduces stochastic-process thinking through two real measured signals. First, a printed graph of "Received Power (moving average) vs traveled distance" is used to show a strongly fluctuating highway radio measurement. The speaker notes that the vertical axis is normalized power in decibels, that the max-to-min variation is roughly a factor of 100, and that repeating the experiment would likely produce a similar but not identical curve because propagation conditions depend on unrepeatable factors such as other cars and exact road position. Handwritten notes connect this to designing better cell phones and simulating cellular systems. The sheet is then replaced by "Rocking Chair: Total acceleration," a time series recorded from a tablet computer while the speaker sat on a rocking chair. He points out noisy beginning and end segments caused by posture changes, an approximately periodic oscillatory middle, and additional ripples that may be harder to model, suggesting a mixed periodic-plus-fluctuation description. Near the end, he begins listing applications such as games, but the sentence is cut off. This introductory segment on stochastic processes uses real-world sensor data to illustrate the need for signal modeling. It first discusses using accelerometers for game controls and biometry, noting that external motions like a rocking chair must be compensated for. It then presents a practical example of measuring voltage across a resistor, where the expected theoretical thermal noise is overwhelmed by a strong, yet not perfectly periodic, hum signal caused by environmental electromagnetic interference. This introductory segment defines stochastic (or random) processes as probability models for signals varying in time or space. Using graphs of resistor noise, the lecturer illustrates the unpredictability and unrepeatability inherent in such signals. The video lists examples including speech, radio transmission power, accelerometer data, and noise. Crucially, the speaker adds a disclaimer that the course treats these as mathematical models for analysis and simulation, without claiming that true stochastic processes necessarily exist in physical reality. This short introductory clip presents stochastic processes, also called random processes, as probability models for signals or functions in time or space. A static handwritten board lists examples such as speech signals, received power in radio transmission, accelerometer data, and noise signals, grouping them under unpredictability, uncertainty, or unrepeatability. The speaker adds that such models are acceptable when they represent the situation well enough for engineering purposes.
Use the learning inspector for key ideas and moments, or open the reading tabs for the complete notes.
Generated from the video's visuals and explanation; not verbatim speech.
The clip opens on a handwritten page headed "Stochastic Processes / Random Processes," while an overlay identifies the video as an introduction to stochastic processes. The lecturer welcomes viewers to the course and immediately clarifies the terminology: in this setting, "stochastic" and "random" are synonyms.
He then writes the core informal definition beneath the title: "Probability models for signals/functions in time or space." In words, he explains that stochastic or random processes are simply names for probability models applied to signals or functions indexed by time or space. This is presented as a motivational definition before any formal mathematical theory.
Next, the lecturer adds a short list of examples on the same page: speech signal, received power in radio transmission, accelerometer data, and noise signal. He states that these are real-world cases where it may make sense to use stochastic-process models, and he chooses the speech signal as the first illustration.
The page changes to two printed plots, each titled as a recording of the phrase "Stochastic Processes." Both graphs use "Amplitude" on the vertical axis and "Sample index" on the horizontal axis. The lecturer explains that he recorded the phrase multiple times on a computer and is showing two of those recordings as discrete functions of sample index.
Using the displayed waveforms, he argues that the signal does not resemble the neat functions usually encountered in elementary mathematical courses. He says it would be very difficult to write the recording in terms of polynomials or cosines, and even if one could produce such a closed form, it would not be especially useful for understanding the signal.
He then compares Recording no. 1 with Recording no. 2. Although both correspond to the same spoken phrase, the plots are not exactly the same: one appears shifted later relative to the other, and there are additional small local differences in waveform shape. The lecturer uses this mismatch between repeated realizations, together with their perceptual similarity on playback, to motivate a probabilistic description of the signal rather than a single deterministic formula.
The segment opens with two printed waveform plots titled "Stochastic Processes", Recording no. 1 and Recording no. 2. Both have vertical axis "Amplitude" and horizontal axis "Sample index" scaled by . The speaker’s point is that the recordings are perceptually similar but not identical, so they are treated as realizations of a random signal rather than a single fixed curve.
The first application is written as "Compress sound". The reasoning given is that if a model tells us which parts of the signal are relevant and which are not, the representation can be reduced while preserving what sounds important to human ears.
The second application is "Denoise the signal". Here the speaker emphasizes removing small unwanted fluctuations. This is not presented as blind filtering alone; it requires a model of the structure of the signal so that meaningful content can be separated from noise-like variation.
The third application is "Speaker recognition". The example is a security system that asks a person to say "stochastic processes" and then decides whether the voice belongs to the authorized individual. Mathematically, this means extracting specific features from the waveform and making the decision rule robust to variation between different recordings of the same phrase.
The page changes to a telecommunications example. A plot titled "Received Power (moving average) vs traveled distance" shows vertical axis "Normalized received power (dB)" and horizontal axis "Traveled Distance [m]". A sketch labels a tower as "Transm." and a vehicle as "Receiver". The narration describes a fixed transmitter and a moving receiver, with received power recorded as a function of distance traveled, producing another fluctuating signal suitable for stochastic-process analysis.
The clip opens on a printed graph titled "Received Power (moving average) vs traveled distance." The horizontal axis is traveled distance in meters and the vertical axis is normalized received power in decibels. A hand-drawn transmitter and receiver sketch in the corner frames the example as a mobile radio measurement along a highway.
The lecturer points out that the curve fluctuates strongly with distance. Because the vertical scale is in dB, he interprets the visible spread from maximum to minimum as roughly a factor of 100 in power. In other words, the graph is not just wiggling slightly; it represents large multiplicative changes in received signal strength.
He then makes the key stochastic-process point: the displayed trace is only one realization. If the same experiment were repeated, the result would probably look similar in character but would not be identical, since the propagation environment during the measurement cannot be reproduced exactly. Factors such as other cars on the road and the precise vehicle position change from run to run.
Beside the graph, handwritten notes summarize why this matters in applications: "Design better cell phones" and "Simulate cell systems." The spoken explanation matches this, saying that telecom engineers use models of received power for device design and for performance simulation of cellular systems.
The sheet is replaced by a second printed graph titled "Rocking Chair: Total acceleration." Now the axes are acceleration in versus time in seconds. The speaker explains that this data came from a tablet computer held while he sat on a rocking chair, shifting the example from communications measurements to physical motion sensing.
Looking at the trace, he identifies large noisy excursions at the beginning and near the end and attributes them to moments when he changed posture. Away from those transients, the middle portion looks fairly periodic: the signal oscillates repeatedly around a baseline near .
He proposes a mixed description rather than a single clean formula. Part of the record can be modeled by a periodic function, but smaller ripples remain on top of that oscillation, and those ripples may be harder to include in a simple model. This makes the example useful for introducing composite signal structure: transient noise plus periodic motion plus residual fluctuation.
Near the end, he begins to mention application areas for modeling such data, writing "Games" on the page. The sentence is cut off before he can elaborate, so only the start of that application discussion is available in this fragment.
The video begins by showing a printed plot titled 'Rocking Chair: Total acceleration', with acceleration on the y-axis and time on the x-axis. The speaker explains that if an accelerometer is used as a control input for a game, one must compensate for external motions, such as sitting in a rocking chair. The plot visually demonstrates this, showing a baseline acceleration with sudden bursts of high-frequency oscillations corresponding to the chair's movement. The speaker writes 'Games' on the paper to label this application.
Next, the speaker introduces another application for acceleration data: biometry. They write 'Biometry' below 'Games'. The idea is to use the acceleration data from a device, like a tablet, to recognize human movement patterns, such as identifying a walker. This requires analyzing the specific signatures left by different gaits on the acceleration signal.
The speaker concludes this section by stating that both gaming and biometry applications require some form of mathematical model for the signals. Without a model to separate the signal of interest from background noise or unwanted motion, these applications cannot function reliably.
The scene changes to two new printed plots titled 'Voltage over a 100 kΩ resistor - Measurement run 1' and 'run 2'. The speaker describes a 'dull experiment' where they measured the voltage across a 100 kΩ resistor using an oscilloscope. Theoretically, the expected result is a small, random signal due to thermal noise inherent in the resistor.
However, the actual measured signals shown in the plots are much more complicated. They exhibit a very strong, dominant periodic wave, which the speaker identifies as a 'hum signal'. This hum is caused by electromagnetic interference (EMI) from other electronic apparatus in the laboratory environment, which easily overwhelms the tiny thermal noise signal.
Finally, the speaker uses a pen to point out specific features on the plots, emphasizing that while the signal has a strong periodic nature, it is 'not exactly periodic'. There are random variations in the amplitude and shape of the peaks and troughs from cycle to cycle. This illustrates that real-world signals often combine deterministic periodic components with stochastic random variations.
The video opens with two plots showing voltage measurements over a 100 kΩ resistor. The speaker uses these to demonstrate that noise signals are inherently unpredictable. By comparing 'Measurement run 1' and 'Measurement run 2', specifically highlighting different spikes around , the visual evidence supports the claim that exact repetition is impossible. The motivation for studying such signals is threefold: analyzing system stability against noise, simulating noisy environments, and designing algorithms to filter or overcome this noise.
The scene shifts to handwritten notes defining 'Stochastic Processes' (also called 'Random Processes') as probability models for signals or functions evolving in time or space. Four concrete examples are listed: speech signals, received power in radio transmission, accelerometer data, and noise signals. A bracket groups these examples under shared characteristics: unpredictability, uncertainty, or unrepeatability. The speaker emphasizes that unlike deterministic systems, these measurements cannot be perfectly replicated even with identical attempts.
Before concluding, the lecturer issues a significant disclaimer regarding the ontology of these processes. He clarifies his earlier choice of words—saying signals 'could be modeled by' rather than 'are' stochastic processes. He admits uncertainty about whether true randomness exists in the physical world, leaving that metaphysical question to philosophers. The course's focus remains pragmatic: developing mathematical frameworks to model and analyze unpredictable signals, regardless of their fundamental physical nature. The segment ends with the guiding question written on the paper: 'How to model such signals?'
The clip opens on a static handwritten board titled "Stochastic Processes / Random Processes". Directly underneath, the board states: "Probability models for signals/functions in time or space". This is the central introductory idea of the segment: the course is treating time-varying or space-varying signals as objects described by probability.
Below the definition, the board lists examples: "Speech signal", "Received power in radio transmission", "Accelerometer data", and "Noise signal". A brace groups these examples with the words "Unpredictability or Uncertainty or unrepeatability". The visual message is that these are not perfectly deterministic, repeatable quantities; instead, they are modeled probabilistically because of their irregular behavior.
At this point the audio continues mid-thought with the phrase "unrepeatable in some way," matching the board’s emphasis on unpredictability and unrepeatability. The speaker then explains the modeling standard being used: "So as long as our models represent the situation well enough for engineering purposes we're all happy and contented." This clarifies that the goal is practical adequacy for engineering use, not exact reproduction of every detail of the real signal.
The speaker concludes the thought with: "And stochastic processes are in many cases good models." Together with the board text, this frames stochastic processes as a flexible modeling tool for signals whose behavior is uncertain or not exactly repeatable.
The final visible prompt on the board is the question "How to model such signals?" No further mathematical construction appears within this 14-second excerpt; the clip functions as a motivational introduction that sets up the next step of the lecture.
At the beginning of the clip, the lecturer explicitly states that the terms "stochastic" and "random" mean the same thing in this course. The handwritten title "Stochastic Processes / Random Processes" visually reinforces that the two names refer to the same subject.
The lecturer defines stochastic or random processes informally as probability models for signals or functions in time or space. This is written directly on the page as "Probability models for signals/functions in time or space" and serves as the conceptual anchor for the rest of the introduction.
Before entering formal theory, the video lists four examples of signals that can be modeled stochastically: speech signal, received power in radio transmission, accelerometer data, and noise signal. These examples show the breadth of the definition introduced just before them.
The first worked example is a speech recording of the phrase "Stochastic Processes." Two printed plots show amplitude versus sample index for Recording no. 1 and Recording no. 2. The lecturer uses these plots to connect the abstract definition to a concrete measured signal.
The lecturer observes that the speech waveform does not look like a standard textbook function. He says it would be very difficult to express it using polynomials or cosines, and even if such a formula existed, it would not help much. This motivates a probabilistic rather than purely deterministic description.
By comparing the two displayed recordings, the lecturer shows that the same spoken phrase produces different sample-index functions across trials. One recording is slightly later than the other and there are small local differences, yet the recordings still sound quite similar when played back. This variability across repetitions is the key motivation for stochastic modeling in the clip.
Two recordings of the same spoken phrase are shown as waveform plots. They have similar overall burst structure but differ in fine detail, illustrating why repeated observations are modeled as realizations of a stochastic process rather than as one deterministic signal.
The first application is audio compression. If a signal model identifies which components are perceptually relevant and which are less important, the representation can be shortened while preserving the parts that matter to the listener.
The second application is denoising. The speaker explicitly says that removing tiny noise fluctuations requires some model of the structure of the signal, so useful content can be distinguished from unwanted variation.
The third application is speaker recognition in a security system. The system listens for the phrase "stochastic processes", extracts specific features, and must adapt to variation across different utterances of the same phrase.
The second motivating example plots normalized received power in dB against traveled distance in meters for a mobile radio measurement. A fixed transmitter and a moving receiver generate a fluctuating received-power curve, again showing a signal whose randomness is worth modeling.
The first example is a printed plot of normalized received power in dB against traveled distance in meters. The lecturer uses it to show a real signal that varies strongly and irregularly along a route, rather than following a simple smooth law.
Because the vertical axis is in decibels, the visible spread from peak to trough corresponds to a large ratio in actual power. The speaker summarizes this as roughly a factor of 100 from maximum to minimum.
The shown curve is not treated as the unique outcome of the experiment. Rerunning the measurement would likely produce a similar but different trace because the propagation conditions during the original run cannot be reproduced exactly.
The lecture connects the random-looking received-power trace to practical engineering use: models of such signals help design better cell phones and simulate cellular systems for performance evaluation.
The second example records total acceleration from a tablet computer while the speaker sits on a rocking chair. This broadens the idea of stochastic-process data from communication measurements to sensor data indexed by time.
The acceleration plot is analyzed in parts: the beginning and end are noisier because of posture changes, the middle is fairly periodic, and even that periodic part contains smaller ripples that may require extra modeling terms.
At the end of the clip, the lecturer begins listing possible application areas for modeling data like the acceleration record and writes "Games," but the explanation is cut off before completion.
Using an accelerometer for game controls requires compensating for unwanted environmental motions, like sitting in a rocking chair. The total acceleration signal contains both the user's intended movements and the background motion, which must be modeled and filtered out.
Acceleration data can be used for biometry, such as recognizing a person's walking pattern (gait) from the data collected by a device they are holding. This involves modeling the specific acceleration signatures of different movements.
Practical applications of sensor data, whether for control or recognition, require mathematical models of the signals. These models allow systems to isolate relevant information from noise and irrelevant background variations.
When measuring voltage across a resistor, the theoretical expectation is small thermal noise. However, in real environments, external electromagnetic interference (EMI) often produces a much stronger periodic 'hum' signal that dominates the measurement.
Even signals dominated by a periodic source, like mains hum, are not perfectly periodic in practice. Random variations in amplitude and shape occur cycle to cycle, making them stochastic processes rather than pure deterministic waves.
Stochastic processes, interchangeably referred to as random processes, are defined as probability models used to describe signals or functions that vary over time or space. This definition establishes the mathematical framework for analyzing systems where outcomes are not strictly determined by initial conditions but involve elements of chance or unpredictability.
Explore conditions, steps and evidence. Supplementary explanations are labeled separately from content shown in the video.
Handwritten title at the top of the page reads "Stochastic Processes".
The lecturer says, "Hello and welcome to the course on stochastic processes."
Stochastic Processes
Name of the course topic; in this video treated as a synonym for random processes.
Course topic name
Handwritten title also includes "Random Processes" separated by a slash from "Stochastic Processes".
The lecturer says, "Actually the term stochastic and random are synonyms, they mean the same thing."
Random Processes
Alternative name for stochastic processes; explicitly stated to be synonymous with "stochastic processes" in this video.
Course topic name
Handwritten line below the title reads "Probability models for signals/functions in time or space".
The lecturer says it is "just a name for probability models for signals or functions in time or space."
Probability models
General class of mathematical descriptions used here for signals or functions indexed by time or space.
Conceptual label for stochastic/random processes
The written definition includes the phrase "signals/functions".
The lecturer says "signals or functions in time or space."
signals/functions
Objects being modeled; the video treats them as time- or space-dependent quantities.
Objects of modeling
The written definition ends with "in time or space".
The lecturer repeats "in time or space."
time or space
Index domains mentioned for the modeled signals or functions.
Index set description
Both printed plots have vertical axis label "Amplitude".
Amplitude
Vertical-axis quantity of the plotted speech recordings.
Signal value axis
The upper plot has horizontal axis label "Sample index" with tick labels 0.4, 0.6, 0.8, 1, 1.2, 1.4, 1.6, 1.8 and a scale marker ×.
The lecturer says he plotted the recording "as a function of the sample index."
The lower plot shows the same numeric ticks but the × marker is not clearly visible in every frame.
Sample index
Horizontal-axis variable indexing discrete samples of the recorded speech signal.
Discrete time/sample axis
Upper plot title reads "Stochastic Processes", Recording no. 1.
Recording no. 1
Label of the first displayed speech waveform example.
Example identifier
Lower plot title reads "Stochastic Processes", Recording no. 2.
Recording no. 2
Label of the second displayed speech waveform example.
Example identifier
Both waveform plots have the vertical axis label "Amplitude".
Amplitude
Vertical-axis quantity of the two audio waveform plots; the displayed values range approximately from -0.3 to 0.3.
Displayed roughly on [-0.3, 0.3] in the two plots.
The horizontal axis is labeled "Sample index" with tick labels 0.4, 0.6, 0.8, 1, 1.2, 1.4, 1.6, 1.8 and a scale marker "×".
Sample index
Horizontal-axis variable for the two recordings, measured in sample number units scaled by .
Visible range approximately [0.4, 1.8] × .
The second plot’s vertical axis reads "Normalized received power (dB)" with ticks from 0 down to -16.
Normalized received power (dB)
Vertical-axis quantity in the telecommunications example, expressed in decibels after normalization.
Visible range approximately [-16, 0] dB.
The lecturer states that "stochastic" and "random" are synonyms and mean the same thing.
The title writes "Stochastic Processes / Random Processes" together on one line.
In this introductory segment, the lecturer explicitly identifies the two names as equivalent terminology for the same subject area.
Applies to the course terminology used in this video.
The lecturer says stochastic or random processes are "just a name for probability models for signals or functions in time or space."
He writes "Probability models for signals/functions in time or space" under the title.
The video gives an informal definition: stochastic processes are probability models for signals or functions indexed by time or space. This is presented as motivation before formal theory.
Informal introductory definition rather than a fully formal measure-theoretic definition.
The modeled objects are signals or functions.
The index domain mentioned is time or space.
The page lists "Examples:" followed by "Speech signal", "Received power in radio transmission", "Accelerometer data", and "Noise signal".
The lecturer says the video will cover these four examples.
Before entering mathematical theory, the lecturer introduces four real-world signal examples that can be modeled as stochastic processes: speech signals, received power in radio transmission, accelerometer data, and noise signals.
These are motivating examples only in this clip; detailed treatment beyond the speech example is not shown within the 120-second segment.
Speaker says the two recordings are "very similar to my ears" but contain variations that may be less relevant perceptually.
Two printed waveform plots titled "Stochastic Processes", Recording no. 1 and Recording no. 2 are shown side by side for comparison.
The clip introduces stochastic processes through repeated realizations of the same spoken phrase. The two waveforms are visually similar in envelope and burst structure, yet not identical sample-by-sample. This motivates treating such signals as random processes rather than fixed deterministic curves: one studies both the recurring structure and the variation across realizations.
The examples are discrete-time sampled audio signals.
The two recordings correspond to the same spoken phrase, "Stochastic Processes".
Speaker says engineers may want to "compress the sound" using information about what is relevant and what is not in the signal.
Handwritten bullet appears: "Compress sound".
Compression is presented as the first engineering use case. If a model identifies which parts of a signal are perceptually important and which are less relevant, the representation can be reduced while preserving the aspects that matter to human hearing.
The signal is an audio recording.
The goal is to retain perceptually relevant structure while reducing representation size.
Speaker says, "Secondly, we may want to denoise the signal," to remove tiny bits of noise in various periods.
Handwritten bullet appears: "Denoise the signal".
Denoising is introduced as removing small unwanted fluctuations from the recording. The speaker explicitly states that this also requires a model of the structure of the signal, so the useful component can be distinguished from noise-like variation.
The observed signal contains both structured content and small unwanted fluctuations.
A structural model is needed to separate signal from noise.
Speaker says, "Finally, the example I would like to mention here is speaker recognition," then describes a security system that asks someone to utter "stochastic processes" and decides whether it is the right person.
Handwritten bullet appears: "Speaker recognition".
The pen points back and forth between the two waveform plots while the speaker discusses matching features across recordings.
Speaker recognition is framed as extracting specific features from a spoken recording and comparing them against expected features for an authorized person. The key mathematical point emphasized is adaptation to variation: the same phrase produces different waveforms, so the decision rule must tolerate realization-to-realization differences.
The input is a spoken audio signal.
Recognition relies on specific features in the signal.
The system must adapt to variations between recordings of the same phrase.
Speaker says, "Our second example is from telecommunications engineering," and describes a transmitter on a mast and a receiver on a van driving on the highway.
A new plot titled "Received Power (moving average) vs traveled distance" is shown.
Handwritten sketch labels include "Transm." near a tower and "Receiver" near a vehicle.
Footer text reads: "Data from: Diversity Measurement Campaign, Telia/FAT, 1998".
The exact university name in the audio is not fully reliable from the available evidence; the visible footer instead identifies Telia/FAT 1998 data.
The second motivating example shifts from audio to wireless communications. A transmitted radio signal is received while the receiver moves, and the recorded quantity is normalized received power as a function of traveled distance. The plotted curve fluctuates strongly around a broader trend, illustrating another type of stochastic signal.
The setting is a mobile radio measurement.
The horizontal variable is traveled distance in meters.
The vertical variable is normalized received power in dB.
A printed graph titled "Received Power (moving average) vs traveled distance" shows a highly irregular blue trace across 0--5000 m.
The speaker says, "And this is the result we get. The signal fluctuates along the traveled distance. It fluctuates quite a lot."
The speaker adds, "This is normalized power in decibels, so what we have here is roughly a factor 100 from the max to the min of the power."
The video presents measured received power along a highway as a concrete example of a signal that varies strongly with position. The plotted quantity is normalized received power in decibels versus traveled distance, and the speaker emphasizes that the curve fluctuates substantially rather than following a simple deterministic trend.
The example is based on a measurement campaign along a highway.
The vertical axis is normalized received power in dB.
The horizontal axis is traveled distance in meters.
The speaker says, "In fact, if we were to run the experiment again, most likely we would get something that looked similar but not the same."
He continues, "The particular realization of this measurement here is determined by the propagation conditions during the measurements, and this we can't repeat because it may depend on the other cars on the road, the exact position on the road, etc."
The lecture uses the received-power trace to motivate stochastic modeling: the exact curve shown is only one realization of the measurement, because the propagation environment during the experiment cannot be reproduced exactly. This introduces the idea that real-world signals may be modeled probabilistically rather than as fixed deterministic functions.
The same experiment is assumed to be rerun under similar but not identical conditions.
The environment includes variables such as other cars and exact road position.
Handwritten notes appear beside the first plot: "Design better cell phones" and "Simulate cell systems".
The speaker says, "Telecom engineer would be interested in such models of these the received power in order to design better cell phone systems or to simulate cell phone systems, perhaps do performance simulations."
The video explains why one would model received power statistically: telecom engineers can use such models to design better cellular devices and to simulate cellular systems for performance evaluation. The method is not computational here; it is a conceptual modeling workflow linking measured random signals to engineering design and simulation.
The signal under consideration is received power in a cellular or mobile-radio context.
The goal is system design or performance simulation.
A new printed graph appears titled "Rocking Chair: Total acceleration" with axes "Acceleration []" and "Time [s]".
The speaker says, "So for this third example, I recorded the total acceleration of a tablet computer that I had in my hand while sitting on a rocking chair."
The lecture introduces another real measured signal: total acceleration recorded by a tablet computer while the speaker sits on a rocking chair. This example broadens the notion of stochastic processes beyond communications data to physical motion measurements over time.
The measured quantity is total acceleration.
The independent variable is time.
The data source is a tablet computer held while sitting on a rocking chair.
The lecturer says, "Actually the term stochastic and random are synonyms, they mean the same thing."
Within this video, "stochastic" and "random" are used to mean the same thing.
The claim is made in the context of naming the course topic "stochastic processes" / "random processes".
Universal within the terminology used in this lecture segment.
The lecturer says stochastic processes are "just a name for probability models for signals or functions in time or space."
The same wording is written on the page.
Stochastic or random processes are probability models for signals or functions in time or space.
This is an introductory, informal characterization rather than a formal theorem.
General descriptive statement about the subject matter.
The lecturer says the first and second recordings are "not exactly the same" even though he said the same word in roughly the same way.
Two printed waveforms labeled Recording no. 1 and Recording no. 2 are shown side by side on the page.
Two recordings of the same spoken phrase are not exactly the same as functions of sample index, despite sounding similar when played back.
The comparison is between the two displayed recordings of the phrase "Stochastic Processes".
Existential comparison of the two shown recordings.
Speaker states the recordings are similar to his ears but contain variations.
The two waveform plots share similar burst envelopes but differ in fine structure and tail behavior.
Two recordings of the same spoken phrase can be perceptually similar while still differing in their sample-level waveform structure.
The recordings are of the same phrase, "Stochastic Processes".
The signals are viewed as sampled waveforms over the same index range.
For the two displayed realizations.
Speaker says compression uses information on what is relevant and what is not in the signal.
Speaker says denoising needs "some kind of model of the structure of the signal".
To compress or denoise such recordings, one needs a model that distinguishes relevant structure from less relevant variation or noise.
The signal is an audio recording with both meaningful structure and variation.
The task is either compression or denoising.
For the audio-processing tasks discussed in this segment.
Speaker says the system must look for specific features in the signal and "you have to adapt for these variations that you may have in the signal".
The pen alternates between the two waveform plots while discussing deciding whether it is the right person.
A speaker-recognition system based on spoken phrases must accommodate variation between different recordings of the same phrase.
Recognition is performed from audio features.
Different utterances of the same phrase produce different waveforms.
For the security-system example described.
The speaker states that rerunning the experiment would most likely produce something similar but not the same, because the realization depends on unrepeatable propagation conditions.
If the received-power experiment along the highway were repeated, the resulting curve would most likely look similar but would not be identical to the displayed realization.
The experiment is repeated in the same general setting.
Propagation conditions are not exactly reproducible.
Environmental factors such as other cars and exact road position vary.
Most likely; applies to a repeated run of the same experiment.
The speaker says the data is noisy at the beginning and end due to posture changes, "but otherwise it's fairly periodic" and that it oscillates.
The central portion of the acceleration plot shows repeated up-down structure around a baseline near .
The claim is qualitative and based on visual inspection rather than a formal periodicity test.
Apart from the noisier beginning and end associated with posture changes, the recorded total acceleration is fairly periodic and oscillatory.
The plotted data comes from the rocking-chair experiment.
The initial and final segments correspond to posture changes.
Qualitative statement about the middle portion of the time series.
It is not exactly periodic. You see here this spike here is much less here and there are other variations this particular part of the signal looks different from this part of the signal.
Pen circling specific peaks and troughs in the plotted signals to show variations.
The measured signal containing electromagnetic interference is not exactly periodic because its amplitude and shape vary across different cycles.
The signal is a measurement of voltage over a resistor in a lab environment.
There exist variations in the signal's spikes and parts.
The lecturer explains that he recorded the words "stochastic processes" and plotted them against sample index, then compares two recordings.
Two printed amplitude-versus-sample-index plots are shown and pointed to with a pen.
Start from a concrete real-world signal: a spoken phrase recorded on a computer.
The lecturer explicitly says he recorded the words "stochastic processes".
Represent the recording as a discrete function of sample index, plotting amplitude versus sample index.
The printed graphs are labeled with "Amplitude" on the vertical axis and "Sample index" on the horizontal axis.
Observe that the resulting waveform does not resemble a simple textbook function such as a polynomial or cosine.
The lecturer says it would be very difficult to write it up in terms of polynomials or cosines.
Compare two separate recordings of the same spoken phrase and note that the plotted signals are not exactly the same.
The lecturer points out differences between Recording no. 1 and Recording no. 2, including timing and small local variations.
Conclude that a deterministic single-formula description is neither easy nor useful here, motivating a probabilistic model instead.
The lecturer says even if one could write such a formula, it would not help much, and uses the variability across recordings as motivation for stochastic processes.
Repeated real-world signals can vary from trial to trial while remaining qualitatively similar, which motivates modeling them as stochastic processes rather than fixed elementary functions.
The two printed plots show similar overall burst structure but different fine details and different apparent onset positions.
The lecturer says, "This one is a bit later than this one," and notes small differences "here and there."
Exact numerical shift between the recordings is not stated.
Some fine visual differences are indicated by pointing rather than by explicit formulas.
Identify the upper plot as Recording no. 1 and the lower plot as Recording no. 2.
The plot titles are printed above each graph.
Note that both plots share the same axes and similar overall envelope structure.
Both use Amplitude versus Sample index and display several speech bursts in the same general order.
Observe that the lower recording appears shifted later relative to the upper one.
The lecturer explicitly says one is a bit later than the other while pointing across the plots.
Observe additional local differences in the waveform details between the two recordings.
The lecturer remarks that there are small differences here and there.
The two recordings are similar in global shape but not identical as sample-index functions.
The speaker enumerates three reasons engineers may model such recordings: compress the sound, denoise the signal, and speaker recognition.
Three handwritten bullets are added in sequence on the right side of the page.
Start from two recordings of the same phrase that are similar overall but differ in detail.
Directly supported by the opening comparison of the two waveform plots and the speaker’s statement about variations.
Recognize that some variations may be less relevant perceptually, so a model can separate important structure from incidental fluctuation.
Stated by the speaker when introducing why engineers may want to model such recordings.
Use that separation for compression: keep what matters and reduce what does not.
Explicitly stated as the first application, accompanied by the written bullet "Compress sound".
Use a structural model for denoising: identify and remove tiny unwanted fluctuations.
Explicitly stated as the second application, accompanied by the written bullet "Denoise the signal".
Use feature extraction plus tolerance to variation for speaker recognition in a security system.
Explicitly stated as the third application, with the written bullet "Speaker recognition" and verbal explanation about adapting to variations.
The variability seen across the two audio realizations motivates stochastic modeling because it underlies compression, denoising, and speaker-recognition tasks.
Speaker describes a transmitter mounted on a mast and a receiver mounted on a van driving on the highway.
The plot title is "Received Power (moving average) vs traveled distance".
The axes are "Normalized received power (dB)" and "Traveled Distance [m]".
Introduce a fixed transmitter and a moving receiver in a highway scenario.
Stated verbally and reinforced by the hand-drawn tower and vehicle sketch.
Record the received signal power while the receiver travels.
The speaker says the signal power was recorded as a function of traveled distance.
Plot normalized received power against traveled distance.
Directly read from the axis labels and title of the displayed graph.
Observe strong local fluctuations around a broader trend in the received power curve.
Visible in the plotted trace, which oscillates sharply over the distance interval.
The telecommunications example yields another stochastic signal, now indexed by distance rather than time samples of speech.
The speaker says, "This is normalized power in decibels, so what we have here is roughly a factor 100 from the max to the min of the power."
The first plot's y-axis spans roughly from 0 dB down to about -16 dB, with visible deep troughs near the lower end.
The spoken "roughly a factor 100" is an approximation tied to the visible extrema rather than a formally derived exact ratio.
The vertical axis is identified as normalized received power in decibels.
Stated directly by the speaker and visible on the plot label.
The visible span from near the maximum to the deepest minima is on the order of tens of decibels; the speaker summarizes this as corresponding to a large multiplicative change in power.
Based on the plotted range and the speaker's verbal interpretation of the graph.
Using the decibel-to-power relation, a change of about 20 dB corresponds to a power ratio of about 100, matching the speaker's statement that the max-to-min variation is roughly a factor of 100.
Standard definition of decibels for power ratios; this step makes explicit the mathematical basis behind the spoken approximation.
The lecturer's statement that the received power varies by roughly a factor of 100 from maximum to minimum is consistent with interpreting the dB-scale fluctuations on the graph.
The speaker distinguishes noisy beginning/end segments from a fairly periodic middle and mentions additional ripples that may be harder to model.
The acceleration plot shows large transient bursts near the start and around 30 s, with smaller repeated oscillations in between.
The decomposition into periodic part plus ripples is qualitative; no explicit formula is written on screen.
Identify the beginning and end of the record as noisier regions associated with posture changes.
Stated by the speaker and visible as larger irregular excursions in the plot.
Treat the remaining middle portion as approximately periodic because it visibly oscillates in a repeating pattern.
Based on the speaker's description and the regular wave-like structure in the central region of the graph.
Model the middle-section signal as a periodic function plus smaller residual ripples that are more difficult to capture.
This expresses the speaker's verbal suggestion that a periodic function can model part of the data but additional ripples remain.
The rocking-chair acceleration example supports a mixed modeling viewpoint: some portions are transient noise, some are approximately periodic, and residual fine structure may require extra modeling terms.
I made a very dull experiment... measure the voltage across this resistor... Theoretically I should be able to measure a small noise signal due to thermal noise... but the signal looked much more complicated like this... what we can see here is that we have a very strong periodic signal with some random variations on it... what this is is a hum signal due to electromagnetic noise
Start with the theoretical expectation of measuring only small thermal noise from the resistor.
Standard physical expectation for a passive component measurement.
Observe the actual measured signal, which is much more complicated and features a strong periodic component with random variations.
Direct visual inspection of the plotted data.
Conclude that the dominant periodic component is a hum signal caused by electromagnetic noise from other lab apparatus, not the expected thermal noise.
Matching the observed periodic nature of the signal with known sources of environmental interference.
The measured signal is dominated by environmental electromagnetic interference rather than the expected internal thermal noise.
A printed page shows two plots titled "Stochastic Processes", Recording no. 1 and Recording no. 2, with axes "Amplitude" and "Sample index".
The lecturer says he recorded the words "stochastic processes" and plotted them as a function of the sample index.
No numerical formula for the signal is given.
Only two of a larger number of recordings are shown.
Use two recorded speech waveforms of the same phrase to motivate why a real-world signal may be modeled probabilistically rather than by one fixed elementary formula.
The recorded phrase is "Stochastic Processes".
Two recordings are displayed as amplitude-versus-sample-index plots.
The lecturer states he repeated the recording several times but shows only two plots.
Show that repeated measurements of the same spoken event differ, suggesting a stochastic-process model.
Record the spoken phrase on a computer.
Stated directly by the lecturer.
Plot amplitude against sample index for each recording.
Visible from the axis labels and stated in the audio.
Compare the two resulting waveforms.
The lecturer visually contrasts Recording no. 1 and Recording no. 2 with the pen.
Note that the waveforms are not exactly the same even though the same word was spoken similarly.
Explicitly stated in the audio and supported by the differing plots.
Note that despite differences, the recordings sound quite similar when played back.
Explicitly stated near the end of the clip.
The example demonstrates that the same real-world speech event produces similar but non-identical sample-index functions, motivating a probability model for such signals.
Verification is qualitative within the clip: the two printed plots visibly differ in timing and fine structure, while the lecturer states they sound similar on playback.
Two printed waveform plots titled "Stochastic Processes", Recording no. 1 and Recording no. 2.
Speaker refers to the recordings and later says the phrase uttered is "stochastic processes".
Handwritten list: "Compress sound", "Denoise the signal", "Speaker recognition".
Given two recordings of the same spoken phrase, explain why they motivate stochastic-process modeling and what engineering tasks can use such a model.
Two waveform plots of the same phrase are displayed.
The recordings are similar to the ear but not identical.
The horizontal axis is sample index and the vertical axis is amplitude.
Identify the modeling need and list the applications mentioned in the clip.
Compare the two waveforms visually and note shared envelope structure with differing fine details.
Direct observation from the two plots and the speaker’s opening comment.
Conclude that the signal should be modeled as a random process with recurring structure plus variation.
This is the conceptual point of the introduction to stochastic processes in the clip.
List compression as the first application, using relevance information to reduce the signal representation.
Explicitly stated by the speaker and written on the page.
List denoising as the second application, requiring a model of signal structure to remove small noise fluctuations.
Explicitly stated by the speaker and written on the page.
List speaker recognition as the third application, where features must be matched despite variation between utterances.
Explicitly stated by the speaker and written on the page.
The two audio recordings motivate stochastic modeling because they are similar yet variable; the applications named are compression, denoising, and speaker recognition.
The answer matches the three handwritten bullets and the speaker’s verbal enumeration in the same interval.
Plot titled "Received Power (moving average) vs traveled distance" with axes "Normalized received power (dB)" and "Traveled Distance [m]".
Speaker describes a transmitter on a mast and a receiver on a van driving on the highway.
Footer: "Data from: Diversity Measurement Campaign, Telia/FAT, 1998".
The exact institution named in the audio is not fully secure from the available evidence.
Describe the second motivating example and identify the plotted quantities.
A radio transmitter is mounted on a mast.
A receiver is mounted on a van driving on the highway.
The displayed graph plots received power against traveled distance.
Explain the setup and read off the variables shown in the plot.
Identify the physical setup from the narration and sketch: fixed transmitter, moving receiver.
Supported by both audio and the hand-drawn labels "Transm." and "Receiver".
Read the horizontal axis as traveled distance in meters.
Directly visible on the plot.
Read the vertical axis as normalized received power in dB.
Directly visible on the plot.
Note that the curve fluctuates strongly over distance, illustrating a stochastic signal in a communication context.
Visible in the plotted trace and consistent with the speaker’s framing of a second example.
The second example is a mobile radio measurement where normalized received power (dB) is plotted versus traveled distance (m), showing large fluctuations along the route.
This matches the plot title, axis labels, footer source line, and the speaker’s description of transmitter and receiver motion.
Printed graph titled "Received Power (moving average) vs traveled distance" with handwritten transmitter/receiver sketch and notes "Design better cell phones" and "Simulate cell systems".
The speaker describes the signal fluctuating along traveled distance, being normalized power in decibels, varying by roughly a factor of 100, and being useful for telecom modeling.
Interpret a measured received-power curve along a highway and explain why it motivates stochastic modeling.
The plot shows normalized received power in dB versus traveled distance in meters.
The curve fluctuates strongly along the route.
The exact realization depends on propagation conditions such as other cars and exact road position.
Describe the observed signal behavior and connect it to engineering uses of probabilistic models.
Read the axes and title to identify the quantity as received power versus distance.
Directly visible on the printed graph.
Note that the trace fluctuates greatly rather than decreasing smoothly.
The speaker explicitly says the signal fluctuates quite a lot, and the graph shows many sharp peaks and troughs.
Translate the dB-scale variation into an approximate multiplicative power variation as discussed by the speaker.
The speaker states "roughly a factor 100 from the max to the min"; the decibel relation supplies the quantitative meaning.
Explain that repeating the experiment would likely give a similar but not identical curve because the propagation environment is not reproducible.
Stated directly in the audio.
Connect the model to applications in designing better cell phones and simulating cellular systems.
Supported by both the spoken explanation and the handwritten notes on the sheet.
The received-power trace is a strongly fluctuating, non-repeatable realization whose statistical modeling is useful for cellular-system design and simulation.
The interpretation matches the visible graph labels, the handwritten application notes, and the speaker's explicit remarks about fluctuation, dB scale, irreproducibility, and telecom use.
Printed graph titled "Rocking Chair: Total acceleration" with axes "Acceleration []" and "Time [s]".
The speaker says he recorded total acceleration of a tablet computer while sitting on a rocking chair, notes noise at the beginning and end from posture changes, and says the rest is fairly periodic with ripples.
The exact numerical amplitude of the periodic component is not stated verbally; only the graph scale is visible.
Analyze a recorded total-acceleration time series from a rocking chair and describe its main structural features.
The data are total acceleration versus time.
The beginning and end look noisier.
The middle portion oscillates repeatedly.
There are additional small ripples on top of the main oscillation.
Characterize the signal and suggest how it might be modeled.
Identify the measured quantity and experimental setup from the title and narration.
Both the printed title and the speaker's description establish that the data are total acceleration from a tablet on a rocking chair.
Point out the noisy intervals at the beginning and end and attribute them to posture changes.
The speaker explicitly links those regions to times when he changed posture in the chair.
Observe that the remaining middle section is fairly periodic and oscillatory.
Visible in the repeated wave pattern and stated by the speaker.
Suggest a model consisting of a periodic function plus residual ripples.
This follows the speaker's statement that part of the data could be modeled by a periodic function but ripples remain harder to include.
The acceleration record combines transient noisy segments with an approximately periodic oscillation plus smaller residual ripples, suggesting a hybrid periodic-plus-fluctuation model.
The conclusion aligns with the visible plot structure and the speaker's verbal segmentation of the time series.
compensate for the motion of the rocking chair... if I want to be able to play the game while sitting in the rocking chair I need to somehow compensate for this effect
Plot titled 'Rocking Chair: Total acceleration' showing a baseline around with two large bursts of high-frequency oscillations.
How to use an accelerometer for game control while sitting in a rocking chair?
An accelerometer provides total acceleration data.
The user is sitting in a rocking chair, introducing periodic motion.
Isolate the game control input from the chair's motion.
Identify the acceleration signal caused by the rocking chair's motion.
The chair's movement creates a distinct, often periodic, acceleration pattern.
Subtract or filter out the identified chair motion signal from the total acceleration data.
This leaves the acceleration related to the user's intentional movements for game control.
By modeling and compensating for the rocking chair's acceleration, the device can accurately track game inputs.
The resulting signal should reflect only the user's hand/body movements intended for the game.
measure the voltage across this resistor using this particular oscilloscope probe... Theoretically I should be able to measure a small noise signal due to thermal noise... but the signal looked much more complicated... very strong periodic signal... hum signal due to electromagnetic noise
Two plots showing 'Voltage over a 100 kΩ resistor' with a strong, noisy periodic wave.
What does the voltage signal across a 100 kΩ resistor look like when measured in a typical lab?
A 100 kΩ resistor.
An oscilloscope for measuring voltage.
A lab environment with other electronic apparatus.
Observe and identify the components of the measured voltage signal.
Set up the oscilloscope to measure the voltage across the 100 kΩ resistor.
Standard procedure for electrical measurement.
Record the signal and observe a strong periodic wave with random variations, rather than just small thermal noise.
Direct observation of the plotted data.
Identify the periodic component as a hum signal from electromagnetic interference in the lab.
The frequency and amplitude match typical mains hum and environmental noise.
The measured signal is dominated by a periodic hum from electromagnetic interference, masking the expected thermal noise.
Moving the setup to a shielded environment or changing the grounding would alter the periodic component.
Two printed graphs titled 'Voltage over a 100 kΩ resistor - Measurement run 1' and 'Measurement run 2'. Axes are Voltage [V] and Time [s]. Blue pen circles highlight different noise spikes at approximately in both runs.
Speaker discusses modeling such signals to study stability, simulate systems with noise, and provide algorithms to overcome noise.
Analyze two measurement runs of voltage over a 100 kΩ resistor to understand noise characteristics.
Graph 1: Voltage over a 100 kΩ resistor - Measurement run 1
Graph 2: Voltage over a 100 kΩ resistor - Measurement run 2
Identify reasons for modeling such signals.
Observe the two graphs showing voltage fluctuations over time.
Visual inspection of the provided plots.
Note that the specific noise spikes (circled in blue) differ between the two runs, demonstrating unrepeatability.
Comparison of the circled regions on both graphs.
Conclude that such signals are modeled to study system stability towards noise, simulate noisy systems, and develop noise-overcoming algorithms.
Audio narration accompanying the visual analysis.
The signals are unpredictable/unrepeatable; modeling them helps in studying stability, simulation, and algorithm design for noise mitigation.
The visual difference in the circled noise spikes confirms the audio claim of unrepeatability.
Handwritten list of examples: Speech signal, Received power in radio transmission, Accelerometer data, Noise signal. Bracketed by 'Unpredictability or Uncertainty or Unrepeatability'.
Speaker lists these as examples of signals that may be modeled as random processes and discusses the philosophical distinction between real-world phenomena and mathematical models.
Provide examples of signals that can be modeled as stochastic processes.
Handwritten notes on paper
List specific physical signals fitting the definition.
Identify 'Speech signal' from the handwritten list.
Direct reading of the text.
Identify 'Received power in radio transmission' from the handwritten list.
Direct reading of the text.
Identify 'Accelerometer data' from the handwritten list.
Direct reading of the text.
Identify 'Noise signal' from the handwritten list.
Direct reading of the text.
The examples are Speech signal, Received power in radio transmission, Accelerometer data, and Noise signal.
All four items are clearly written on the paper shown in the video.
The board explicitly lists four example signals under "Examples:".
The same examples are bracketed with "Unpredictability or Uncertainty or unrepeatability".
Identify the examples shown for signals that motivate stochastic processes.
Speech signal
Received power in radio transmission
Accelerometer data
Noise signal
List the examples displayed on the board and their common feature.
Read the examples directly from the board.
The examples are written explicitly under the heading "Examples:".
Note the brace grouping them with "Unpredictability or Uncertainty or unrepeatability".
The board visually associates all four examples with those three properties.
The examples are speech signal, received power in radio transmission, accelerometer data, and noise signal; they are grouped as unpredictable, uncertain, or unrepeatable.
The answer matches the visible handwritten list and brace annotation on the board.
A white sheet on a wooden desk shows the handwritten heading "Stochastic Processes / Random Processes".
A title overlay briefly appears reading "STOCHASTIC PROCESSES INTRODUCTION" and "TROELS PEDERSEN, AALBORG UNIVERSITY".
Handwritten page heading
Title overlay text
Presenter's hand and pen
The overlay title appears and then disappears, leaving the handwritten heading visible.
The handwritten heading remains at the top of the page.
The segment opens by identifying the topic as stochastic/random processes and signaling that this is an introduction.
The presenter writes a new line beneath the heading.
The completed line reads "Probability models for signals/functions in time or space".
Pen
Handwritten definition line
A blank area under the heading is filled with the definition sentence.
The heading stays unchanged above the new line.
The visual act of writing reinforces the spoken informal definition of stochastic processes.
The presenter writes a list under the heading "Examples:".
The list contains "Speech signal", "Received power in radio transmission", "Accelerometer data", and "Noise signal".
Handwritten list
Pen
Four bullet-style examples are added one after another below the definition.
The earlier title and definition remain visible above the list.
The page shifts from abstract definition to concrete application domains.
The page changes to two printed plots titled "Stochastic Processes", Recording no. 1 and Recording no. 2.
The pen points along the waveforms and across corresponding regions of the two plots.
The exact amount of horizontal shift between the recordings is not quantified on screen.
Upper waveform plot
Lower waveform plot
Axis labels "Amplitude" and "Sample index"
Pen pointer
The presenter moves the pen between the two plots to indicate differences in timing and local waveform shape.
Both plots use the same type of axes and represent the same spoken phrase.
The visual comparison supports the claim that repeated realizations of the same signal are similar but not identical.
Two stacked waveform plots titled "Stochastic Processes", Recording no. 1 and Recording no. 2 occupy the left page.
A pen points to corresponding bursts in the upper and lower traces, then moves to the blank right page.
Upper waveform plot
Lower waveform plot
Pen
Right-hand writing area
The pen first indicates the lower plot, then the upper plot, then shifts right to write notes.
Three bullets are added sequentially: "Compress sound", "Denoise the signal", "Speaker recognition".
Both plots use the same axis labels: Amplitude and Sample index.
Both titles refer to the same phrase, "Stochastic Processes".
The visual comparison establishes that the two recordings share overall structure but differ in detail, motivating a probabilistic description of signal variation.
A new page shows the plot "Received Power (moving average) vs traveled distance".
At the top right there is a hand-drawn tower labeled "Transm." and a vehicle labeled "Receiver" on a road line.
The pen points to the tower, then to the vehicle, then traces along the plotted curve.
Received-power plot
Tower sketch
Vehicle sketch
Road line
Pen
The page changes from audio waveforms to a distance-versus-power graph.
The pen links the sketch elements to the plotted data by pointing at transmitter, receiver, and curve.
The plot title and axis labels remain fixed throughout this interval.
The footer source line remains visible at the bottom.
The sketch maps the physical measurement setup onto the graph: a fixed transmitter and moving receiver produce a received-power trace indexed by traveled distance.
A printed sheet fills the frame showing the title "Received Power (moving average) vs traveled distance", a jagged blue curve, axis labels, and a hand-drawn transmitter/receiver sketch in the upper right.
A hand with a pen points along the curve and axis regions while the speaker discusses fluctuation and dB scale.
Printed line graph
Title text
Y-axis label "Normalized received power [dB]"
X-axis label "Traveled Distance [m]"
Hand-drawn transmitter tower
Hand-drawn receiver vehicle
Pen tip
The pen moves across different parts of the curve to indicate peaks, troughs, and axis ranges.
Attention shifts from the overall trace to the vertical dB scale and then back to the horizontal distance axis.
The underlying printed graph remains the same throughout this interval.
The title and axis labels stay visible.
The visual focus supports the spoken claim that received power varies irregularly with traveled distance and that the vertical scale is logarithmic in the sense of decibels.
Two handwritten bullet points become visible on the right side of the first sheet: "Design better cell phones" and "Simulate cell systems".
The speaker mentions telecom engineers using such models for design and simulation.
The exact moment when each word was written is not fully visible; only the finished notes are clearly seen in this interval.
First printed graph
Handwritten note "Design better cell phones"
Handwritten note "Simulate cell systems"
Pen
New text is added to the right side of the sheet.
The pen points toward the notes as the speaker names the applications.
The received-power graph itself does not change.
The transmitter/receiver sketch remains in place.
The added notes visually summarize the practical motivation for modeling the random received-power signal.
The first sheet is removed and replaced by a new printed sheet titled "Rocking Chair: Total acceleration".
The speaker introduces this as the third example and describes recording tablet acceleration while sitting on a rocking chair.
First sheet leaving frame
Second printed graph
Title "Rocking Chair: Total acceleration"
Y-axis label "Acceleration []"
X-axis label "Time [s]"
The whole visual context changes from a distance-indexed communication signal to a time-indexed motion signal.
The new graph shows a different waveform shape with large bursts near the ends and oscillations in the middle.
The lecture format remains a paper-based visual explanation with spoken commentary.
This transition broadens the concept of stochastic-process examples from wireless measurements to physical sensor data.
A hand points to the noisy beginning, the smoother oscillatory middle, and the later burst around 30 s on the acceleration graph.
Near the end of the clip, the pen writes a bullet point and the word "Games" below the graph.
The speaker discusses posture changes, periodicity, ripples, and possible relevance to games.
The sentence begun after "Games" is cut off before completion, so the full intended application list is not available.
Acceleration graph
Pen tip
Handwritten bullet point
Word "Games"
The pen traces the early noisy segment, then the middle oscillatory region, then the later large excursion.
A new handwritten note "Games" is added beneath the graph.
The graph title and axes remain unchanged.
The overall shape of the acceleration trace stays fixed while only the pointing and writing change.
The visual annotation reinforces the spoken decomposition into transient noise, periodic motion, and residual ripples, and begins to connect such data to application domains such as games.
A printed plot titled 'Rocking Chair: Total acceleration' is shown. The y-axis is 'Acceleration []' (0-12) and x-axis is 'Time [s]' (0-35). The blue line shows a relatively stable baseline around 10, with two distinct periods of high-amplitude, high-frequency oscillations (around and ). Handwritten text 'Games' and later 'Biometry' are added.
Printed graph
Blue signal line
Handwritten text 'Games'
Handwritten text 'Biometry'
Pen
Pen points to the oscillating parts of the signal.
Handwrites 'Games' and 'Biometry' on the paper.
The axes labels and scales remain constant.
The underlying printed signal does not change.
The plot illustrates real-world acceleration data that contains both a steady component (gravity) and dynamic components (motion), which need modeling for applications like gaming or biometry.
Two printed plots titled 'Voltage over a 100 kΩ resistor - Measurement run 1' and 'run 2'. Y-axis is 'Voltage [V]' (-0.02 to 0.04) and x-axis is 'Time [s]' (0 to 0.09). Both show a strong, noisy periodic wave. A physical oscilloscope probe is held near the plots. The pen circles specific peaks and troughs to highlight variations.
Two printed graphs
Blue signal lines
Oscilloscope probe
Pen
Pen points to different parts of the waveforms.
Pen circles specific peaks and troughs to show they are not identical.
Physical probe is introduced into the frame.
The axes labels and scales remain constant.
The overall periodic nature of the two runs is similar.
The plots demonstrate that a simple theoretical expectation (thermal noise) is often overwhelmed by complex real-world factors (electromagnetic interference), resulting in signals that are periodic but not strictly so due to random variations.
The lecturer says the signal looks strange, does not resemble functions usually found in mathematical courses, and would be very difficult to write using polynomials or cosines.
One might expect a usable signal model to be written directly as a simple polynomial, cosine, or similar closed-form expression.
The video argues that for this speech example such a formula would be difficult to obtain and, even if available, would not necessarily be helpful.
The lecturer explicitly notes that the first and second recordings are not exactly the same even though the same word was said in roughly the same way.
One might assume that saying the same word twice should produce exactly the same signal.
The displayed recordings differ in timing and fine waveform details, although they remain perceptually similar.
Speaker says the recordings are very similar to his ears but there are variations in the signals.
The two waveform plots are not identical despite similar titles and envelopes.
One might think that if two recordings sound similar, their waveforms should be the same.
The clip shows that perceptual similarity can coexist with substantial sample-level variation, which is exactly why a stochastic-process model is useful.
Speaker says denoising requires "some kind of model of the structure of the signal".
One might assume denoising only means applying a generic filter without understanding the signal.
The speaker explicitly ties denoising to a structural model that distinguishes meaningful signal content from noise-like fluctuations.
Speaker says the system must look for specific features and adapt for variations in the signal.
One might imagine that recognizing a speaker from a phrase means matching the recorded waveform exactly.
The example stresses feature extraction and adaptation to variation across different utterances of the same phrase.
The speaker explicitly contrasts getting something "similar but not the same" upon repetition with the impossibility of repeating the exact propagation conditions.
One might think the displayed received-power curve is the exact result that would be obtained every time the experiment is repeated.
The video stresses that the shown curve is only one realization; rerunning the experiment would likely produce a similar but different trace because the propagation environment cannot be reproduced exactly.
The speaker says the data are noisy at the beginning and end due to posture changes, and even the periodic-looking part contains ripples that may be harder to model.
The acceleration plot visibly contains large transient bursts and fine irregularities in addition to the middle oscillation.
Because the middle of the rocking-chair plot oscillates, one might conclude that the entire dataset is simply periodic.
The video distinguishes noisy transient sections from the approximately periodic middle and further notes residual ripples, so the full signal is not purely periodic.
Theoretically I should be able to measure a small noise signal due to thermal noise in the resistor. But the signal looked much more complicated like this... what this is is a hum signal due to electromagnetic noise
One might assume that measuring voltage across a resistor will only yield the theoretical thermal noise inherent to the component.
In practical environments, external factors like electromagnetic interference from other apparatus often produce much stronger signals (like a periodic hum) that dominate the measurement.
It is not exactly periodic. You see here this spike here is much less here and there are other variations...
Pen circling different peaks on the plotted waveform to show amplitude and shape differences.
A signal with a clear repeating pattern might be mistakenly thought to be perfectly periodic.
Real-world signals, even those dominated by a periodic source like mains hum, contain random variations and noise, making them 'not exactly periodic' as the amplitude and exact shape fluctuate cycle to cycle.
Speaker says, 'I deliberately avoided saying we will look at real world examples of stochastic processes. This is because I don't know if a stochastic process actually exists in real life... This question I think is better left to the philosophers to answer.'
Believing that stochastic processes definitely exist as fundamental realities in the physical world rather than just being useful mathematical models.
The speaker clarifies that it is unknown if true randomness exists in reality; the course focuses on how to model signals mathematically, regardless of their ontological status.
The speaker says, "So as long as our models represent the situation well enough for engineering purposes we're all happy and contented."
One might think a stochastic model must reproduce the real signal exactly to be useful.
The video states that a model is acceptable if it represents the situation well enough for engineering purposes.
The lecturer first equates "stochastic" and "random", then defines the subject as probability models for signals or functions.
The terminology equivalence sets up the informal definition: both names refer to the same class of probability models discussed in the video.
The written definition is followed by a list of examples on the same page.
The lecturer says there are many examples and then lists four before diving into mathematical theory.
The examples instantiate the informal definition by showing concrete signal types that can be modeled as stochastic processes.
"Speech signal" appears in the written example list, and the next page shows two speech recordings.
The lecturer says the first example is a speech signal and then discusses the plotted recordings.
The speech-signal example is the first specific case developed from the general list of motivating examples.
The spoken explanation of the two recordings directly supports the motivational argument for stochastic modeling.
The two printed plots are used as visual evidence in that argument.
The intuitive derivation relies on the speech example as its concrete evidence.
The compression discussion follows directly from the observation of relevant versus less relevant variations in the recordings.
The idea of modeling relevant structure in noisy or variable recordings is applied to audio compression.
The denoising bullet is introduced as the second reason to model the signal structure.
Structural signal modeling is applied to separating desired content from small unwanted fluctuations.
The speaker-recognition example is presented as the final application of modeling variable recordings.
Variability across repeated utterances motivates feature-based recognition that adapts to differences between realizations.
The speaker explicitly calls the telecommunications case "Our second example".
The page changes from two audio waveforms to a received-power-versus-distance plot.
Both examples motivate stochastic processes, but the first is indexed by audio sample number and the second by traveled distance in a mobile radio setting.
The discussion of non-reproducible propagation conditions directly follows the presentation of the received-power graph.
Understanding the received-power trace as a measured fluctuating signal is needed before appreciating why its exact realization is treated as random.
The speaker moves from irreproducible realizations to why telecom engineers want models of received power.
Handwritten notes list design and simulation applications beside the first graph.
The probabilistic view of the received-power realization motivates its use in cellular-system design and simulation.
After introducing the acceleration example, the speaker analyzes its noisy and periodic parts and suggests modeling implications.
The rocking-chair acceleration example contains the more specific observation that the signal mixes transient noise, periodic oscillation, and residual ripples.
The clip presents two different printed examples: one indexed by traveled distance and one indexed by time.
The speaker treats both as examples of data that can be modeled, though in different application contexts.
This comparison is an analyst-level synthesis rather than an explicit statement by the speaker.
Both examples illustrate real measured signals suitable for stochastic modeling, but the first is a spatial communication measurement while the second is a temporal motion measurement.
The lecturer defines stochastic/random processes in words during this interval.
The definition is also written on the page.
The lecturer explicitly says the terms are synonyms.
The lecturer says the speech signal would be hard to express with polynomials or cosines and that such a formula would not help much.
The lecturer compares Recording no. 1 and Recording no. 2 and notes they are not exactly the same but sound similar.
The two printed waveforms are shown side by side.
The lecturer says it is a good idea to look at real-world examples before diving into mathematical theory.
The page lists four examples.
Opening comparison of two recordings of the same phrase.
Two waveform plots shown together.
Speaker explains compressing the sound using relevance information.
Bullet "Compress sound".
Speaker says denoising needs a model of the structure of the signal.
Bullet "Denoise the signal".
Speaker describes a security system and adapting to variations in the signal.
Bullet "Speaker recognition" and pen pointing between the two waveforms.
Plot title and axis labels for received power versus traveled distance.
Description of transmitter on mast and receiver on van.
The speaker says the dB-scale plot corresponds to roughly a factor 100 from max to min.
The speaker explains that rerunning the experiment would give something similar but not the same because propagation conditions cannot be repeated.
Covered · Opening title, course identification, and statement that stochastic and random are synonyms.
Covered · Written and spoken informal definition of stochastic processes as probability models for signals/functions in time or space.
Covered · List of four motivating examples written on the page and named aloud.
Covered · Speech-signal example with two plotted recordings, visual comparison, and motivational conclusion.
Covered · Opening comparison of the two waveform recordings and the motivation for stochastic modeling.
Covered · First application: compressing sound using relevance information in the signal.
Covered · Second application: denoising the signal and the need for a structural model.
Covered · Third application: speaker recognition from spoken phrases with adaptation to variation.
Covered · Brief page transition with no new mathematical content beyond the switch to the next example.
Covered · Telecommunications example with received power plotted against traveled distance.
Covered · First example: received power versus traveled distance, fluctuation discussion, dB interpretation, irreproducibility, and telecom applications.
Covered · Second example: transition to rocking-chair total acceleration, identification of noisy transients, approximately periodic middle, residual ripples, and the beginning of an application note "Games".
Covered · Discussion of accelerometer applications in games and biometry, requiring signal models, illustrated by a rocking chair acceleration plot.
Covered · Analysis of a measured resistor voltage signal, contrasting expected thermal noise with observed electromagnetic interference and discussing its non-exact periodicity.
Covered · Analysis of resistor noise graphs and motivations for modeling.
Covered · Introduction of formal definition, list of examples, shared properties, and philosophical disclaimer.
Covered · The entire 14-second clip consists of a static introductory board plus spoken motivation about stochastic models and engineering adequacy.
Reviewed subject paths
Stochastic processes, also called random processes, are informally defined as probability models for signals or functions in time or space. This definition serves as a motivational introduction before formal mathematical theory is developed.
Conditions: Applies to the introductory terminology used in this video.; The modeled objects are signals or functions.; The index domain mentioned is time or space.
Continue with related videos in the same subject.
YouTube5:38
YouTube10:32
YouTube1:53