[sdiy] Pole Mixing Multi-Mode filter design with SSI2140

Mon May 4 17:20:37 CEST 2026

Thanks Tom, 

One of the things I wish this page did was to allow me to type in exact (fractional) numbers for the filter poles so I can see the small changes in the response. The sliders are fine, but seem to have a very course resolution. This would allow one to see how the filter responds based on smalls changes in accuracy of resistors. I did not see a way of doing that - is there one?

Thanks

Jeff

Sent by voice-transit conductor

> On May 4, 2026, at 3:09 AM, Tom Wiltshire <tom at electricdruid.net> wrote:
> 
> Just a reminder of this excellent page for experimenting with pole-mixing filter combinations:
> 
> https://rnd.expeditionelectronics.com/Diy/Polemixing
> 
> This can show you how sensitive different modes are to tolerances. Shift a mix value very slightly and see what happens. Sometimes it's a lot. 1% precision is probably the *minimum* required for certain responses.
> 
> I don't think I agree about Mattias' comment about the 4-pole HP. At least, I can't get the sim to match what he describes, so at least in theory it shouldn't do that.
> 
> Tom
> 
> 
>> On 4 May 2026, at 08:15, Mattias Rickardsson <mr at analogue.org <mailto:mr at analogue.org>> wrote:
>> 
>> Hi, 
>> Pole mixing is fascinating, but from my experiences from employing it on a four-pole cascade lowpass filter I'd recommend to think twice. Some drawbacks are not obvious from the start - in the order of decreasing obviousness:
>> 
>> - Precision. Especially the resistors need a certain precision, which can be less fun if looking for those deep notches or if you need matching stereo pairs or synth voices.
>> 
>> - Noise. The more complex pole mix modes rely on mixing many taps very loudly and then cancelling out most of their signals except for the desired ones. The catch is that noise is not cancelled out, and adds together with lots of gain. Noisefloor is then much higher on some modes.
>> 
>> - Resonance. The pole-mixed filter slopes are correct when the resonance is turned down, but not when you turn it up. Resonant peaks can then appear in less desirable ways. Also the stopband suffers: The 4-pole HP mix for instance is then not 24 dB/oct anymore since an increasing resonance pulls up a flatter response. I guess it would need some resonance switching scheme or resonance-dependent pole mixing to get this right.
>> 
>> So in general I'd recommend to stay away from it and create the esoteric 4-pole responses in other ways, unless you want to try it for fun. :-)
>> 
>> But the few pole-mixing possibilities on 2-pole filters work very well!
>> 
>> /mr
>> 
>> PS: I've seen some DC offset issues as well!
>> 
>> Den fre 1 maj 2026 17:12Jeff Whitman via Synth-diy <synth-diy at synth-diy.org <mailto:synth-diy at synth-diy.org>> skrev:
>>> 
>>> 
>>> Hello there,
>>> 
>>> I’ve been working on an analog synth project for several years using Sound Semiconductor parts. Their data sheets and design notes have been incredibly helpful, especially for those of us who are not primarily analog designers. I’m a firmware engineer by profession - now retired, with an EE background from a long time ago, so I understand the basics but would really appreciate input from people with more analog design experience.
>>> 
>>> For this project, I’ve built a “voice” of three oscillator boards using the SSI2131, a filter board using the SSI2140 in the suggested pole-mixing configuration to create a 16-mode multimode filter, a low-pass filter using the SSI2144, and linear-controlled VCAs using the SSI2164.
>>> 
>>> My questions are specifically around using the SSI2140 for pole mixing. From what I can tell, this design is very similar to the filter used in the Matrix-12, and I’ve spent some time studying that schematic. I’ve also reviewed the AN701 filter design document on the Sound Semiconductor website. I’m including my schematic in PDF form in the hope that some of the very smart analog folks here can offer suggestions about something I’m seeing.
>>> 
>>> I’ve been using Claude.ai <http://claude.ai/> to help analyze each of my filter configurations and determine whether they are behaving as expected. I’ve also been doing sound comparisons against the Arturia Matrix-12 V modeled synth. One thing I’ve found is that I’m getting DC offset that appears to be coming from the pole outputs of the SSI2140. Depending on the pole-mixing resistor configuration selected for a given filter mode, I’m seeing significant DC offset.
>>> 
>>> I’ve been feeding Claude scope shots to help analyze this, and so far its interpretation seems pretty consistent with what I’m measuring. As a side note, this has been a very useful application of AI for troubleshooting. For the pole-mixing network, I’m using 1% resistors and also using 1% capacitors, as recommended in the data sheet and AN701.
>>> 
>>> Here is the table Claude generated from my measurements:
>>> 
>>> Mode    Mean DC Offset
>>> LP1     -12.5 mV
>>> LP2     -40 mV
>>> LP3     -30 mV
>>> LP4     -30 mV
>>> HP1     -61.9 mV
>>> HP2     -131.1 mV
>>> HP3     -284.4 mV
>>> HP4     -583.6 mV
>>> Claude’s current explanation is that the increasing weighting factors in the pole-mixing network are directly amplifying the DC component.
>>> 
>>> Things that appear to have been ruled out:
>>> 
>>> Power supply asymmetry (-11.93 V vs. +12.02 V) — likely negligible
>>> Input signal DC component — input is about +117 mV, while U1A shows only -8.33 mV offset
>>> TL074 op-amp offset voltage — too small to explain this magnitude
>>> Unity-gain buffers — not populated - I have jumpers.
>>> The suggested fix is to add four 1 µF film capacitors, such as WIMA MKS2 parts, on the SSI2140 OUT1–OUT4 pins, placed between the OUT pins and the 15k mixing resistors. The idea is that this would block DC before it enters the pole-mixing network, preventing it from being amplified by U1B regardless of filter mode.
>>> 
>>> I’m curious whether adding these coupling capacitors makes sense, and whether that would be compatible with the intended use of the SSI2140. I’m a little surprised that the data sheet does not discuss the need for them if this is expected behavior. I did find some discussion about this on Mod Wiggler, but I didn’t see a clear conclusion.
>>> 
>>> I’d appreciate any advice, and hopefully I’ll be able to understand it. Also, if you see any other design issues or possible improvements, I’d be grateful for the feedback — pun intended.
>>> 
>>> Thanks,
>>> 
>>> Jeff
>>> 
>>> 
>>> Sent by voice-transit conductor
>>> 
>>> ________________________________________________________
>>> This is the Synth-diy mailing list
>>> Submit email to: Synth-diy at synth-diy.org <mailto:Synth-diy at synth-diy.org>
>>> View archive at: https://synth-diy.org/pipermail/synth-diy/
>>> Check your settings at: https://synth-diy.org/mailman/listinfo/synth-diy
>>> Selling or trading? Use marketplace at synth-diy.org <mailto:marketplace at synth-diy.org>
>> ________________________________________________________
>> This is the Synth-diy mailing list
>> Submit email to: Synth-diy at synth-diy.org <mailto:Synth-diy at synth-diy.org>
>> View archive at: https://synth-diy.org/pipermail/synth-diy/
>> Check your settings at: https://synth-diy.org/mailman/listinfo/synth-diy
>> Selling or trading? Use marketplace at synth-diy.org <mailto:marketplace at synth-diy.org>
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://synth-diy.org/pipermail/synth-diy/attachments/20260504/96d9790e/attachment.htm>