Probably get away with software fpu in this case, not sure how fast the processor but would be surprised if it couldn't keep up. Very good presentation for beginners though, impressive!!!
120 Mhz means at that sample frequency it is little more than 1000 machine code instructions per channel to do the filtering... not so sure it would fit a software fp.
I had tens of hours of digital signal processing and signal filtering lectures, but this is the most intuitive course I've come across. Thank you for the clear explanations.
Nice demo! I did something very similar a few days ago to add some low pass filtering for a speed sensor in a motor control loop firmware, using a jupyter notebook as well to design a simple FIR filter. I actually then used python to directly generate C code containing the array with the filter coefficients.
Just watched the presentation of the two PCB's. I like the dev kit pcb"s and ordered them for delivery tomorrow from RS Components in the UK. 🇬🇧 Thank-you both
The algorithm presented in 54:52 will not work.....the variables must retain the previous information, so they cannot be reset every time they enter in the function. They have to be outside, that is, be global and not local. I think that's it..... the programmers can confirm it or not!
One thing I wish was brought up: layering stages. For instance in the 1khz tone example he showed a tighter filter takes longer to build, would there be an advantage to layering a looser filter with a tight filter to decrease that initial blip while still getting that tight rejection? Or would the loose filter make the tight filter redudndant? If you wanted to reject multiple bands do the two stages add delay independently, or is the delay the same as if they were applied individually? Do you add these stages to do so, or is there a means to merge them into a smaller structure?
Great questions and thoughts @xxportalxx !! Cascading multiple second order sections (one after the other if that is what you mean by layering) is the common approach to implement higher order filters. The big advantage in doing this is isolating the filter poles from interacting with each other leading to higher stability implementations, especially when the poles are close to the frequency axis ("higher Q" resonators). That's the primary advantage, and the delay and performance is predicted by the prototype higher order filter. If we used that approach to isolate two different frequencies as the notch filter demonstrated here, the delay (which is only in vicinity of the notch and creates the start-up transient) is as it would be for each filter individually, as long as they are separated sufficiently to not interact with each other. The deeper details are beyond the scope of this short video or what I could fairly answer in a comment, but it is something I go into as part of the "DSP for Wireless Communications" course referenced in the link (where we have much more time and the right context to explain). That course is coming up next month!
@danboschen9613 Amazing presentation, and thank you for your response. DSP is generally presented as a black box above us mere mortals, very eye opening demonstration that it doesn't have to be out of reach.
Digital signal processing was one of my favorite college courses. The math was tough, but learning how to make practical IIR and FIR filters was a great payoff. Audio is a great way to experiment because you can experience it. Most of my work nowadays is in RF / software defined radio processing which I also find fascinating
some MCU vendors (like ST) have dedicated hardware (FMAC) to do these calculations in fixed point numbers. so the DMA could stream directly to and from the FMAC and the CPU can do other things or sleep. just want to point out how solid this concept is that it is implemented in hardware.
Finally! Someone that can actually explain digital filters so that we mere humans can actually understand.
And enthusiastically does so !
Probably get away with software fpu in this case, not sure how fast the processor but would be surprised if it couldn't keep up. Very good presentation for beginners though, impressive!!!
120 Mhz means at that sample frequency it is little more than 1000 machine code instructions per channel to do the filtering... not so sure it would fit a software fp.
@@robegatt yeah good point, one would have to try to be sure..
I had tens of hours of digital signal processing and signal filtering lectures, but this is the most intuitive course I've come across. Thank you for the clear explanations.
Excellent content, Robert. Thank you so much for sharing.
Nice demo! I did something very similar a few days ago to add some low pass filtering for a speed sensor in a motor control loop firmware, using a jupyter notebook as well to design a simple FIR filter. I actually then used python to directly generate C code containing the array with the filter coefficients.
God bless people like you.
I think that was a great presentation of DSP, and it was great to see your enthusiasm for the subject and teaching. :)
Cool video, the low and hi pass filter?
Cool that digilent PCB, Mouser has it.
Very cool video and Dan is an absolutely amazing teacher. I should know, as I am one of his students :-)
It was very informative. Thanks a lot.
Thank you! This is the best video I have seen on this topic so far!
Just watched the presentation of the two PCB's.
I like the dev kit pcb"s and ordered them for delivery tomorrow from RS Components in the UK. 🇬🇧
Thank-you both
The algorithm presented in 54:52 will not work.....the variables must retain the previous information, so they cannot be reset every time they enter in the function. They have to be outside, that is, be global and not local.
I think that's it..... the programmers can confirm it or not!
@chinametal they are defined as “static variables” which means they retain their state from call to call.
Wish I had my teachers explained this clearly......on the other hand may be I should have paid more attention
Excellent video, thank you guys! 😀
Beautifully explained sir 👏 👌 ❤️
Good dsp presentation thanks
The subject material is very well presented - thank you.
It would be great to have a video delay to match the audio filter delay.
One thing I wish was brought up: layering stages. For instance in the 1khz tone example he showed a tighter filter takes longer to build, would there be an advantage to layering a looser filter with a tight filter to decrease that initial blip while still getting that tight rejection? Or would the loose filter make the tight filter redudndant? If you wanted to reject multiple bands do the two stages add delay independently, or is the delay the same as if they were applied individually? Do you add these stages to do so, or is there a means to merge them into a smaller structure?
Great questions and thoughts @xxportalxx !! Cascading multiple second order sections (one after the other if that is what you mean by layering) is the common approach to implement higher order filters. The big advantage in doing this is isolating the filter poles from interacting with each other leading to higher stability implementations, especially when the poles are close to the frequency axis ("higher Q" resonators). That's the primary advantage, and the delay and performance is predicted by the prototype higher order filter. If we used that approach to isolate two different frequencies as the notch filter demonstrated here, the delay (which is only in vicinity of the notch and creates the start-up transient) is as it would be for each filter individually, as long as they are separated sufficiently to not interact with each other. The deeper details are beyond the scope of this short video or what I could fairly answer in a comment, but it is something I go into as part of the "DSP for Wireless Communications" course referenced in the link (where we have much more time and the right context to explain). That course is coming up next month!
@danboschen9613 Amazing presentation, and thank you for your response. DSP is generally presented as a black box above us mere mortals, very eye opening demonstration that it doesn't have to be out of reach.
So nice thanks sir
Your blog are great and you're great, great, great.
Excelente !!!!
Magnetics design for SMPS please
3rd to view 1st to like the video (>.
The first 2 views were mine (before publishing) so you are the first one to see it after publishing :)
Thank you very much for the video, but I want you to add Turkish subtitles, I am looking forward to your feedback
Digital signal processing was one of my favorite college courses. The math was tough, but learning how to make practical IIR and FIR filters was a great payoff. Audio is a great way to experiment because you can experience it. Most of my work nowadays is in RF / software defined radio processing which I also find fascinating
You can hear the 1 khz tone in the background all the time. Very annoying. Now I need to build a filter before watching this video.
Ha ha! I am 1 KHz tone deaf from all the testing I did
Fantastic
wow
Just amazing!
Thank you very much
Dakujem za velmi uzitocne video
another ... thank you.
I understand there are specialized courses, but it would be great if you could share the demo jupyter notebook. thanks.
Optimized version would not pass coefficients everytime.
the problem of delay you can remove it by using a zero phase shift filters , great video Robert
This is a real time filter and zero-phase filters are non-causal (so will only work with post processing and can’t be applied here).
some MCU vendors (like ST) have dedicated hardware (FMAC) to do these calculations in fixed point numbers. so the DMA could stream directly to and from the FMAC and the CPU can do other things or sleep. just want to point out how solid this concept is that it is implemented in hardware.
Professor was on some BS when asked where his notch filter coefficients came from.