SSI2164 datasheet 18db/oct VCF clipping

From circuitbending to homebrew stompboxes & synths, keep the DIY spirit alive!

Moderators: lisa, luketeaford, Kent, Joe.

Post Reply
ZTX500
Learning to Wiggle
Posts: 14
Joined: Sun Jun 21, 2020 3:27 am

SSI2164 datasheet 18db/oct VCF clipping

Post by ZTX500 » Thu Jun 25, 2020 5:19 pm

I've built the 18db/oct VCF in figure 18 of the SSI2164 datasheet, http://www.soundsemiconductor.com/downl ... asheet.pdf and am getting surprising results - DC inputs greater than about 3V cause the LP output to suddenly diverge to the positive rail, and likewise signals less than -3V likewise hit the negative rail. Is this the expected behavior?

The similar schematic at https://www.njohnson.co.uk/index.php?me ... bsubmenu=1 claims ±5V input levels, but in simulation I see this having similar problems above and below about ±2.5V.

Does anyone have any suggestions for improving the behavior of this filter when it is overdriven?

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1373
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by neil.johnson » Thu Jun 25, 2020 5:53 pm

What happens if you block DC with a series capacitor? The original schematic has an AC-coupled input.

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

ZTX500
Learning to Wiggle
Posts: 14
Joined: Sun Jun 21, 2020 3:27 am

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by ZTX500 » Thu Jun 25, 2020 7:34 pm

The results seem essentially the same with AC coupling. Obviously DC and anything much below the ~16Hz AC-coupling frequency is attenuated, but signals in the pass band of the LP output (i.e. above ~16Hz and somewhat smaller than cutoff frequency) exhibits the same problem.

The way I have understood it is to assume feedback is acting to bring the difference between the input voltages to zero for all op-amps, and work out when that cannot be true - this is when the output hits the rails. For any signals far below the cutoff frequency,
  • HP, UBP and LPB are close to zero and there is negligible current flowing into or out of the 2164 amplifiers,
  • therefore the output of the feedback op-amp (IC4A in version on your website) is at -V_in in order to sink all of the current coming from the input.
  • Since UBP=LBP=0, no current flows in the resonance part of the feedback network, so the voltage on the back-to-back zeners must be (R12/R11)V_in, where R12/R11=1.42 in the website version, and 1.1 in the SSI2164 datasheet version
  • But the voltage across the zeners cannot be more than ~3.3V, so feedback stops working when (R12/R11)V_in=±3.3V, leading to maximum V_in of about ±3V (SSI2164 datasheet) or ±2.5V (website version).
Possibly I would get the saturation behaviour I want by taking the LP output from (a buffered version of) the voltage across the zeners, but I was hoping to avoid this, partly because it would require a respin of my PCB and rebuild, and partly because I guess it would compromise the accuracy of some of the combined filter responses (notch etc.)

Ideally I'd like to avoid any of the opamps ever hitting the rails, even for very hot input signals. For this, possibly just a soft limiter on the input is the best option, though I thought that limiting within the filter itself would likely sound better.... just not sure if there is any way of doing this...

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1373
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by neil.johnson » Fri Jun 26, 2020 10:58 am

Umm, the gain around IC4A is R11/R12. In the original, this is 0.7 to the feedback stops around 4.7V. In the SSI2164 datasheet the gain is around 0.91, so limiting at about 3.6V.

Oh, and welcome to Muffs.

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1373
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by neil.johnson » Fri Jun 26, 2020 11:00 am

Also note that pole mixing as presented in the "new" version doesn't quite work on the old version due to the slightly skewed damping control. The old filter still sounds good, although at 12 years old I'm surprised it is still getting some attention.

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1373
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by neil.johnson » Fri Jun 26, 2020 11:09 am

Anyway, if I get some time over the weekend or next week I'll dig out the test board and run some tests.

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

ZTX500
Learning to Wiggle
Posts: 14
Joined: Sun Jun 21, 2020 3:27 am

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by ZTX500 » Fri Jun 26, 2020 7:18 pm

Thanks Neil,

The attached ltspice screenshot (modelling the new SSI2164 schematic) shows the behaviour I am referring to. All 2164 stages are omitted here, corresponding to unity current gain, and so a ~22kHz cutoff frequency and minimum resonance. Input is a ±5V 1Hz sine wave - though results are similar at 1kHz.
opa1678_18db_vcf_investigation_3pole.png

I get similar results from the real thing, though in this trace I've changed the zeners for around 4.9V types, so the effect only occurs at >11Vpp signals. The problem with this mod is that with the greater amplitudes, high resonance settings cause clipping fairly readily - even if I increase the effect of the resonance amplitude limiting part of the circuit (with the back-to-back 1N4148s).
NewFile0.png
You do not have the required permissions to view the files attached to this post.

User avatar
oldenjon
Ultra Wiggler
Posts: 760
Joined: Sun Feb 26, 2012 1:28 am
Location: Sacramento, CA

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by oldenjon » Sat Jun 27, 2020 12:04 am

The zeners might not be necessary. OPA167x has phase reversal protection. Have you tried using TL07x as per the datasheet schematic?
"We all dance by ourselves, that's why we only play electronic music"

http://synthscalledbeasts.tumblr.com/

ZTX500
Learning to Wiggle
Posts: 14
Joined: Sun Jun 21, 2020 3:27 am

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by ZTX500 » Sat Jun 27, 2020 8:51 am

You're right, they may not be necessary with OPA167x. I've not tried the TL07x - the datasheet does recommend OPA167x, even though it's TL07x in the schematic.

I had hoped (wrongly, I think) that the zeners also provided some saturation when the filter is overdriven, but maybe this isn't actually desirable - I can imagine that nonlinearity doesn't help with the 'pole mixing' accuracy.

I still don't fully understand even the linear behaviour of the filter. In the SSI2164 datasheet the transfer function is given as H(s) = F³/(s³+D(Fs²+F²s)+F³), but as far as I can see this is only true if the feedback has gain 1, which would require the 30k resistor in the datasheet figure 18 (setting the gain of the feedback path op-amp) to actually be 36.3k. Theoretically, whether it is 30k or 36.3k doesn't seem to have a huge effect on where the poles of the filter are, but I'd be interested to know why the 30k choice was made...

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1373
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by neil.johnson » Sat Jun 27, 2020 9:05 am

Once a filter saturates then all bets are off since you are operating the devices outside of the normal conditions. Opamps banging against the rails means you have lost control - you don't even have any guarantees about what voltage they will put out, how long they will take to recover, etc.

The original design (from my website) dates back to around 2001. Some of the values were derived empirically to tame it. When I revisited the design for SSI I took the opportunity to scale down the resistors to more sane values.

I'm curious: is this a DIY project, or are you working towards a commercial product?

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

ZTX500
Learning to Wiggle
Posts: 14
Joined: Sun Jun 21, 2020 3:27 am

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by ZTX500 » Sat Jun 27, 2020 11:56 am

It's just a DIY project - but my interest is more in learning about electronics than music, thus all the questions about how it works! I'm starting to get to grips with the theory, but with filters in particular it seems that maths only gets you so far and at some point empirical judgement and expertise (which I don't have!) takes over.

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1373
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by neil.johnson » Sat Jun 27, 2020 1:10 pm

That's a very laudible goal! But the number of text books on filter theory suggest that, while the theory is fine to a point, it does not model the real world (a lot of the advanced text books focus on dealing with the imperfections of the real world). For example, as far as I know no-one in the audio world compensates for op-amp phase shifts, no-one uses Akerberg Mossberg compensation - it's an extra op-amp per integrator so extra cost and complexity. Instead we futz around with the feedback circuit to keep it stable from things like Q enhancement. And how many designers have done sensitivity analysis to work out which components are the critical ones and so what tolerance their design needs? Or, put another way, who cares?

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

User avatar
guest
Super Deluxe Wiggler
Posts: 4233
Joined: Mon Aug 19, 2013 11:49 am

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by guest » Sun Jun 28, 2020 3:08 pm

out of curiousity, what is the intent for the zeners at the output? i understand diode compression on a positive feedback pathway -> as the amplitude gets bigger, the positive feedback reduces, and you gain some stability. but those zeners are in the negative feedback pathway, so as the amplitude gets bigger, you lose negative feedback, and the output saturates.
openmusiclabs.com

ZTX500
Learning to Wiggle
Posts: 14
Joined: Sun Jun 21, 2020 3:27 am

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by ZTX500 » Mon Jun 29, 2020 8:03 am

guest wrote:
Sun Jun 28, 2020 3:08 pm
out of curiousity, what is the intent for the zeners at the output? i understand diode compression on a positive feedback pathway -> as the amplitude gets bigger, the positive feedback reduces, and you gain some stability. but those zeners are in the negative feedback pathway, so as the amplitude gets bigger, you lose negative feedback, and the output saturates.
This is essentially my confusion with this schematic. In the SSI2164 datasheet, Neil (I think) writes "Two back-to-back 2V7 zener diodes limit the main loop feedback signal to ±3.3V. This is necessary due to the TL072’s phase reversal behavior which would otherwise lock up the filter loop.".

Without paying too much attention to the circuit initially, I had assumed that the zeners would do this by reducing gain at high amplitudes, but as you say they do the opposite. Maybe they do something to help the circuit out of lockup state if TL072 phase reversal takes place? (my tests have been with OPA1678/9 op-amps, which don't have phase reversal, so I haven't looked into this at all).

User avatar
guest
Super Deluxe Wiggler
Posts: 4233
Joined: Mon Aug 19, 2013 11:49 am

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by guest » Mon Jun 29, 2020 8:30 am

thanks, thats good to know. i wonder which opamp was going into phase reversal. it looks like they will all be within the common mode range for all conditions. have you tried it without the zeners?
openmusiclabs.com

ZTX500
Learning to Wiggle
Posts: 14
Joined: Sun Jun 21, 2020 3:27 am

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by ZTX500 » Wed Jul 01, 2020 4:56 am

guest wrote:
Mon Jun 29, 2020 8:30 am
have you tried it without the zeners?
I've just done so (both in reality and simulation). As far as I can see the zeners have three effects:
  1. Large amplitude inputs at f << f_c are magnified and eventually saturate on the output rails due to diode compression in the negative feedback path - as discussed above.
  2. The filter resonance (at f=f_c) is decreased when there is a large amplitude DC (or f<<f_c) signal present, as the zeners cause a relative decrease of LP feedback amplitude relative to HBP+LBP feedback, and this pushes the poles back towards the stable region. The effect of this is to suppress the amplitide of resonant 'ringing' on (e.g.) high amplitude low frequency square or saw inputs, where the input signal immediately following the discontinuity is of high amplitude. This is particularly noticeable on an offset square wave input (toggling between 0V and 2.3V) where the resonance is much reduced about the 2.3V input state:
    withzeners.png
    Compare with the case without zeners:
    withoutzeners.png
    The other form of nonlinearity in the circuit (the back-to-back 1N4148s) is AC-coupled so produces a resonance amplitude that does not depend on DC (or f<<f_c) offset.
  3. The zeners prevent the self-resonant oscillation amplitude at very low frequencies (a few tens of Hz) from becoming very large and saturating the op-amps. I'm assuming this is due to the resonance reduction effect of the previous point, but it's not clear to me why the circuit without zeners seems to be more resonant at these low frequencies. (I don't think it's the C_x capacitor, which acts at ~20kHz not ~20Hz)
Effect 1 is the one I had noticed previously, but I suspect 2 and perhaps 3 are the reasons the zeners are in the schematic.

For any others building this filter - using the recommended class A biasing of the 2164 (7.5kΩ r_bias) I was seeing 200-600mV DC (!) of control voltage feedthrough at the LP output, and small but noticeable noise. Switching to class AB improved both of these considerably. With nonlinear elements in the filter anyway I'm not too concerned about the increase in distortion that results. Something seems wrong with that feedthrough voltage... even in class A I think it should be <10mV...
You do not have the required permissions to view the files attached to this post.

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1373
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Re: SSI2164 datasheet 18db/oct VCF clipping

Post by neil.johnson » Wed Jul 01, 2020 7:55 am

1. As you have found out, avoid large inputs.
2. Filters do not have "negative" feedback - they have feedback, and sometimes feedforward.
3. This particular filter implements a summed-feedback topology. If you start summing outputs to get other transfer functions then you have summed-feedback summed-output topology.
4. When the resonance is turned right up to self-oscillation the outer feedback path with the zeners closes the loop and keeps it oscillating - at which point the zeners stabilise the amplitude. There are other ways of doing this, including using another gain cell and a level detector similar to what is done in Wien bridge oscillators and would generate a very pure sine wave. But that would have defeated the original goal.
5. When self-oscillating any signal input doesn't really make sense (you're not AM or FMing the signal, you're just injecting another signal into the oscillator). You'll most likely introduce distortion, or overdrive one of the stages, or some other undesirable outcome.

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

Post Reply

Return to “Music Tech DIY”