Second Order Biquad filters

[PAk-CatchFire]

I see it is possible now to export second-order sections format / Biquad IIR in SOS format coefficients

It would be interesting to also set the resolution to float32 (and also float64) format, since I have seen some issues because of that in the past, so this would be great tfor checking the stability of the filter, instead of just passing digits.

Thank you again.
Best Regards

[chipmuenk]

Yes, I've started working on that, although with focus on fixpoint formats and arithmetics. Adding float 32 would not be a big problem.

The big question is: What do you mean by "set the resolution"? Coefficients? Poles/zeros? Or arithmetics, i.e. the handling of summation and multiplication rounding errors?

When regarding arithmetic inaccuracies, the order of calculation in the filter is very important, different filter topologies have different sensitivities - although they have the same ideal system function H(z).

Perhaps you could tell me your exact use case and then we can discuss what would make sense to implement in pyfda.

[PAk-CatchFire]

I have designed in the past a very narrow filter for a TI DSP, and when I was drawing the H(f) response in Matlab, it was also fine.

However, the DSP only had " float32 format" enabled by default, and I was seeing some unexpected transitories in realtime. After some investigation, this croping in the float format made the filter response to change and not converge because of the last digits of the coefficients (they were slightly different from the float64).

So yes, I meant coefficients. It would be nice tyo be able to design the filter for both float32 and float64, and check the true responses.

[chipmuenk]

Exporting the coefficients in float32 format is not too hard, I've got this more or less implemented in the development branch. I'm planning to provide an 'quantize' option where float64 coefficients and SOS coefficients (both calculated with a 64 bit algorithm) not only can be diplayed and exported with 32 or 64 bits but also saved with 32 bits. That way you would see whether the filter e.g. becomes unstable with lower coefficient length. Is that what you mean by "design the filter for both float32 and float64, and check the true responses."?

[PAk-CatchFire]

Thank you @chipmuenk .
That would be great, that is exactly what I am talking about.

Please, mind that because of float32 resolution (which is much faster to compute on ARMs and DSPs), some decimal digits will change because of resolution.
Most devices use IEEE 754 Single-precision floating-point format.

Would it be possible to see the change on the graphic when choosing/toggling between 32bit or 64bit format?

Regardss

[chipmuenk]

I think that is exactly what I plan to do: Right now, there is an option to display coefficients in various fixpoint formats, providing different views on the coefficients. Only when you press the quantize button (the staircase) are the quantized coefficents stored and the frequency response, P/Z, impulse response ... plots updated. That is what you mean by "see the change on the graphic ", isn't it?

[PAk-CatchFire]

Yes, it is!!
I have checked to generate a current float filter, then quantize it to integer and applychanges and I see the graphics updated.
I understand that now in float you will have both float64 and float32, right?

[PAk-CatchFire]

Hello @chipmuenk.
I wonder if the latest commit includes this functionality

Thanks!!

I think that is exactly what I plan to do: Right now, there is an option to display coefficients in various fixpoint formats, providing different views on the coefficients. Only when you press the quantize button (the staircase) are the quantized coefficents stored and the frequency response, P/Z, impulse response ... plots updated. That is what you mean by "see the change on the graphic ", isn't it?

[PAk-CatchFire]

I think that is exactly what I plan to do: Right now, there is an option to display coefficients in various fixpoint formats, providing different views on the coefficients. Only when you press the quantize button (the staircase) are the quantized coefficents stored and the frequency response, P/Z, impulse response ... plots updated. That is what you mean by "see the change on the graphic ", isn't it?

Hi @chipmuenk
Do you have an estimation for this feature? I really think it would be of great help for the embedded community.
Thanks again!!

[chipmuenk]

Hi Pak, due to lots of private trouble I don't have the time to work on pyfda at the moment. I think I'll find some time to implement that feature in October and create a new release. Please have some patience till then! Regards, Christian Am 26.07.25 um 00:13 schrieb Pak:

*PAk-CatchFire* left a comment (chipmuenk/pyfda#268) <#268 (comment)> I think that is exactly what I plan to do: Right now, there is an option to display coefficients in various fixpoint formats, providing different views on the coefficients. Only when you press the quantize button (the staircase) are the quantized coefficents stored and the frequency response, P/Z, impulse response ... plots updated. That is what you mean by "see the change on the graphic ", isn't it? Hi @chipmuenk <https://github.com/chipmuenk> Do you have an estimation for this feature? I really think it would be of great help for the embedded community. Thanks again!! — Reply to this email directly, view it on GitHub <#268 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABDH55CTJUSHZ5525QDYILD3KKTYHAVCNFSM6AAAAAB2767KJGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTCMRQGUZTCOBZGE>. You are receiving this because you were mentioned.Message ID: ***@***.***>

--------------NMhr9Z893GwO4HzOWVCc3STV Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: 8bit <!DOCTYPE html><html><head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> </head> <body> <p>Hi Pak,</p> <p>due to lots of private trouble I don't have the time to work on pyfda at the moment. I think I'll find some time to implement that feature in October and create a new release. Please have some patience till then!</p> <p>Regards,</p> <p>Christian<br> </p> <div class="moz-cite-prefix">Am 26.07.25 um 00:13 schrieb Pak:<br> </div> <blockquote type="cite" ***@***.***"> <div style="display: flex; flex-wrap: wrap; white-space: pre-wrap; align-items: center; "> <img height="20" width="20" style="border-radius:50%; margin-right: 4px;" decoding="async" src="https://avatars.githubusercontent.com/u/4245130?s=20&amp;v=4" moz-do-not-send="true"><strong>PAk-CatchFire</strong> left a comment <a href="#268 (comment)" moz-do-not-send="true">(chipmuenk/pyfda#268)</a></div> <blockquote> <p dir="auto">I think that is exactly what I plan to do: Right now, there is an option to display coefficients in various fixpoint formats, providing different views on the coefficients. Only when you press the quantize button (the staircase) are the quantized coefficents stored and the frequency response, P/Z, impulse response ... plots updated. That is what you mean by &quot;see the change on the graphic &quot;, isn't it?</p> </blockquote> <p dir="auto">Hi <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/chipmuenk/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/chipmuenk" moz-do-not-send="true"> @chipmuenk</a><br> Do you have an estimation for this feature? I really think it would be of great help for the embedded community.<br> Thanks again!!</p> <p style="font-size:small;-webkit-text-size-adjust:none;color:#666;">—<br> Reply to this email directly, <a href="#268 (comment)" moz-do-not-send="true"> view it on GitHub</a>, or <a href="https://github.com/notifications/unsubscribe-auth/ABDH55CTJUSHZ5525QDYILD3KKTYHAVCNFSM6AAAAAB2767KJGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTCMRQGUZTCOBZGE" moz-do-not-send="true"> unsubscribe</a>.<br> You are receiving this because you were mentioned.<img src="https://github.com/notifications/beacon/ABDH55ANCIT5FFLSF5R5LWT3KKTYHA5CNFSM6AAAAAB2767KJGWGG33NNVSW45C7OR4XAZNMJFZXG5LFINXW23LFNZ2KUY3PNVWWK3TUL5UWJTVZ76E3G.gif" height="1" width="1" alt="" moz-do-not-send="true"><span style="color: transparent; font-size: 0; display: none; visibility: hidden; overflow: hidden; opacity: 0; width: 0; height: 0; max-width: 0; max-height: 0; mso-hide: all">Message ID: <span>&lt;chipmuenk/pyfda/issues/268/3120531891</span><span>@</span><span>github</span><span>.</span><span>com&gt;</span></span></p> <script type="application/ld+json">[ { ***@***.***": "http://schema.org", ***@***.***": "EmailMessage", "potentialAction": { ***@***.***": "ViewAction", "target": "#268 (comment)", "url": "#268 (comment)", "name": "View Issue" }, "description": "View this Issue on GitHub", "publisher": { ***@***.***": "Organization", "name": "GitHub", "url": "https://github.com" } } ]</script> </blockquote> <pre class="moz-signature" cols="72"> </pre> </body> </html>

…

--------------NMhr9Z893GwO4HzOWVCc3STV--

[PAk-CatchFire]

Thank you for your answer.
I totally understand.

On my side, I have been testing and what it would be needed to work in a single float precision is the cascade Biquad format.

Do you think it would be possible to have this implementation?
Best regards