sir is there any specific reason behind using orthonormal signals as the basis function for the purpose of correlation (for the purpose of modulation and demodulation) and sir why is the timing recovery necessary for the correct detection of data?
These videos might help: "How are Complex Baseband Digital Signals Transmitted?" th-cam.com/video/0lkRJgnywkg/w-d-xo.html and "Orthogonal Basis Functions in the Fourier Transform" th-cam.com/video/n2kesLcPY7o/w-d-xo.html and "What is a Matched Filter?" th-cam.com/video/Ci-EjiMJo3I/w-d-xo.html
Nice explanation as always. Thank you so much. Perhaps you can make a video about carrier and symbolclock regeneration. That would be great. Keep up the great teaching.
In general, direct sampling is only possible if you have an Analog-to-Digital converter that can sample the received waveform with high voltage resolution at twice the carrier frequency (although there are sophisticated techniques for so-called "sub-sampling" at lower rates that make use of aliased copies, but these "lower rates" are still well beyond the symbol rates). Then it is necessary to run signal processing algorithms on these high rate digital samples. All of this has implications for cost and power consumption in the receiver.
I think this is showing hard decisions (based on decision boundaries), but there is also soft decision where the point represents the probability that it was originally in that position. Then we apply something like Viterbi which actually says what the transmitted symbols were.
Well, yes and no. The Viterbi algorithm also makes "hard decisions" (although there are Soft-Output versions), and it's not generally referred to as a "detector". The Viterbi algorithm is used as a "decoder" and as an "equalizer". But you're right, there are other "soft output" detectors, and I've now put that topic on my "to do" list, so thanks for the suggestion.
Nice video!! I wonder if there's a coding scheme that increases the likelihood of a symbol being sent if it is nearer the extremities of the constellation diagram, since it has a lower probability of error associated with it?
Good idea. Of course, the implementation aspects of how to encode such a mapping need to be factored in too. And that effect is less when the constellation size grows, and there are fewer "outer edge" symbols and more "inner" symbols.
Hi Professor! Great video. Q: How would the detection decision be physically implemented on HW/SW? I am thinking that it’s a software program with if/else statements for the different boundaries.
The samplers shown at the output of the demodulator are implemented with Analog-to-Digital Converters (ADC), so yes, the Detector is implemented with digital computations. But that doesn't mean it is with _software_ . Often it is implemented in hardware in an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA).
Professor, I am getting the opportunity to pursue my interest in wireless research, and wanted to thank you for your contribution towards that interest! Your explanations (especially of that pesky OFDM cyclic prefix) really show the cleverness in the ideas designed by the field. I was hoping to get your perspective on machine learning's impact in the coming years. I have seen that direct optimization of the physical layer w/ ML is not proving to be earth shattering. Is this your experience, and do you think there is opportunity elsewhere?
Glad you're finding the videos helpful. Machine learning is a powerful tool so there are many possibilities. It's very important to feed a learning algorithm with relevant feature vectors derived from the raw data, in order for it to work well. So it's really a combination of understanding/modelling enough about a system in order to design the feature vectors, and then see if the ML algorithm can handle the remaining nonlinearities. I think there is still quite a bit of experimentation/research to explore along these lines.
Is the order on the website for the videos important or random? It would be nice to have some sort of scheme for what to watch first.
The order of the topics on the website are what I would suggest for learning the material from scratch.
sir is there any specific reason behind using orthonormal signals as the basis function for the purpose of correlation (for the purpose of modulation and demodulation)
and sir why is the timing recovery necessary for the correct detection of data?
These videos might help: "How are Complex Baseband Digital Signals Transmitted?" th-cam.com/video/0lkRJgnywkg/w-d-xo.html and "Orthogonal Basis Functions in the Fourier Transform" th-cam.com/video/n2kesLcPY7o/w-d-xo.html and "What is a Matched Filter?" th-cam.com/video/Ci-EjiMJo3I/w-d-xo.html
Nice explanation as always. Thank you so much. Perhaps you can make a video about carrier and symbolclock regeneration. That would be great. Keep up the great teaching.
Thanks for the suggestion. I'll add it to the list.
crystal clear explanation perfect!
Glad it was helpful!
Sir, why we ise ıQ Signal in place of direct sampling? Can you make a video about quadrature Signal?
In general, direct sampling is only possible if you have an Analog-to-Digital converter that can sample the received waveform with high voltage resolution at twice the carrier frequency (although there are sophisticated techniques for so-called "sub-sampling" at lower rates that make use of aliased copies, but these "lower rates" are still well beyond the symbol rates). Then it is necessary to run signal processing algorithms on these high rate digital samples. All of this has implications for cost and power consumption in the receiver.
I think this is showing hard decisions (based on decision boundaries), but there is also soft decision where the point represents the probability that it was originally in that position. Then we apply something like Viterbi which actually says what the transmitted symbols were.
Well, yes and no. The Viterbi algorithm also makes "hard decisions" (although there are Soft-Output versions), and it's not generally referred to as a "detector". The Viterbi algorithm is used as a "decoder" and as an "equalizer". But you're right, there are other "soft output" detectors, and I've now put that topic on my "to do" list, so thanks for the suggestion.
Nice video!! I wonder if there's a coding scheme that increases the likelihood of a symbol being sent if it is nearer the extremities of the constellation diagram, since it has a lower probability of error associated with it?
Good idea. Of course, the implementation aspects of how to encode such a mapping need to be factored in too. And that effect is less when the constellation size grows, and there are fewer "outer edge" symbols and more "inner" symbols.
Hi Professor! Great video. Q: How would the detection decision be physically implemented on HW/SW? I am thinking that it’s a software program with if/else statements for the different boundaries.
The samplers shown at the output of the demodulator are implemented with Analog-to-Digital Converters (ADC), so yes, the Detector is implemented with digital computations. But that doesn't mean it is with _software_ . Often it is implemented in hardware in an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA).
Professor, I am getting the opportunity to pursue my interest in wireless research, and wanted to thank you for your contribution towards that interest! Your explanations (especially of that pesky OFDM cyclic prefix) really show the cleverness in the ideas designed by the field.
I was hoping to get your perspective on machine learning's impact in the coming years. I have seen that direct optimization of the physical layer w/ ML is not proving to be earth shattering. Is this your experience, and do you think there is opportunity elsewhere?
Glad you're finding the videos helpful. Machine learning is a powerful tool so there are many possibilities. It's very important to feed a learning algorithm with relevant feature vectors derived from the raw data, in order for it to work well. So it's really a combination of understanding/modelling enough about a system in order to design the feature vectors, and then see if the ML algorithm can handle the remaining nonlinearities. I think there is still quite a bit of experimentation/research to explore along these lines.
why is it that we have to know the start and end of the symbol duration before we begin with the detection ??
Perhaps this video will help: "What is a Matched Filter?" th-cam.com/video/Ci-EjiMJo3I/w-d-xo.html
Why would a fading channel have the effect of rotating the constellation?
You might like to watch this video for insights into the answer: "What is Rayleigh Fading?" th-cam.com/video/-FOnYBZ7ZfQ/w-d-xo.html
nice video! love you