Portable two-band Parametric EQ (ESP/Urei 545)

[abbey road d enfer]

Hi @abbey road d enfer , sorry for insisting on this but what is the reason the SVF oscillates when forced to cover a full range spectrum (ie. 40Hz-15KHz). I thought it would be due to particular opamp limitations they 'run out of juice', but I tried with other opamps with very high output specs and bandwidth and same happens. Could you please share with me what the limiting factor is?

The probable reason is that the two halves of the tuning pot don't track closely enough and at some point the loop gain is too high.
Does the circuit oscillate whatever the position of the tuning pot?
40-15k is a ratio of 300+. SVF's filters have a typical range of 1:30, which often uses switching for covering the audio spectrum in two ranges.

 

[domingo]

The probable reason is that the two halves of the tuning pot don't track closely enough and at some point the loop gain is too high.

That's an interesting hipothesis, didn't think of that.

Oscillation occurs only (and always) at low frequency and high gain/Q. Roughly when fq is below 80Hz, Q 6.9 and gain above 7.

If I increase the resistor between the frequency pots and ground oscillation doesn't occur, for the pot ends then at 80Hz. But then I loose range which is what I'm trying to increase and understand where the obstacle is.

Another symptom I have is that, when Q is low and gain high (roughly Q 2 and gain 7), the filter simply collapses and stops filtering, letting the source signal passthrough after a few intermittences. But this is not oscillation, seems like the opamp runs out of power, but other opamps fail at the same point.

 

[abbey road d enfer]

That's an interesting hipothesis, didn't think of that.

Oscillation occurs only (and always) at low frequency and high gain/Q. Roughly when fq is below 80Hz, Q 6.9 and gain above 7.

If I increase the resistor between the frequency pots and ground oscillation doesn't occur, for the pot ends then at 80Hz. But then I loose range which is what I'm trying to increase and understand where the obstacle is.

You are pushing the limits of the design. I guess (would need to make a sim to be more affirmative) the 100uF caps at the bottom of the pots introduce significant phase-shift that impairs stability. Can you try replacing them with larger values (470 or even 1000uF)?
Why don't you use a range switch, that would solve your problem?

 

[domingo]

You hit right on the spot @abbey road d enfer ! I'm very glad that I asked and that you were there with an answer... After increasing the caps to some 220uF (solid) elcos oscillation immediately stopped )) So the problem at low fq and high Q/gain is over. I've been dealing with this problem for quite a while so I am really deeply happy and thankful with your suggestion. And my ears too!

You are pushing the limits of the design.

I'm attached to an equalising practice that I inherited from digital filters (spoiled by!), of which I can hardly get rid of. It normally starts with a band without knowing if lows or highs will be filtered, then as I scan the whole range I create a curve and then move to the next one. In other words I usually don't know if I'll filter highs or lows until I start experimenting and find something. If ranges are separated I easily get trapped. Also, another habit from digital filtering, very often I need to create peak curves that are close to each other, so the shared range between bands should be wide as possible. Switching ranges/caps unfortunately doesn't help me for the intended use/workflow, which involves recording while eq'ing.

Line apart.
If not too much already. I'm still having issues on the opposite corner: when Q and gain are to a wide minimum low, while frequency is to a maximum high (ie., Q 0.4, A -11, 15KHz). Right before reaching the end of any three pots, being the other two already at the end, the signal starts suffering from very fast interruptions; if pushed even further, the filter simply shuts off and stops working. With certain extreme ratios (+180) I sensed that, instead of shutting down, the filter's gain gets sort of inverted, sudently amplifying the whole high to ultra-high range rather than attenuating it. But in less extreme ratios that I'm intending to use, around 140, the filter just just off and an attenuated source signal goes through instead. Would you have any idea on the cause of this? Ground caps don't seem to influence, neither the type of opamp. Maybe the integrators' resistors are too low (R200), but not fully sure that's the cause of it. Maybe unfiltered DC, not sure.

 

[abbey road d enfer]

If not too much already. I'm still having issues on the opposite corner: when Q and gain are to a wide minimum low, while frequency is to a maximum high (ie., Q 0.4, A -11, 15KHz). Right before reaching the end of any three pots, being the other two already at the end, the signal starts suffering from very fast interruptions; if pushed even further, the filter simply shuts off and stops working. With certain extreme ratios (+180) I sensed that, instead of shutting down, the filter's gain gets sort of inverted, sudently amplifying the whole high to ultra-high range rather than attenuating it. But in less extreme ratios that I'm intending to use, around 140, the filter just just off and an attenuated source signal goes through instead.

I don't understand what you mean by "ratio". Ratio is a term taht pertain to compressor/gates, not to EQ's.

 

[domingo]

I don't understand what you mean by "ratio".

I mean the ratio between the highest and lowest frequency of the range covered by a band (ie. 107Hz-15KHz = ratio 140).

 

[sahib]

I have read all of the posts now. I am wondering if increasing the gain (reducing R146 to 2k) range is giving you this headache as Abbey mentions of curves moving and touching unintended frequencies.

 

[abbey road d enfer]

It's very hard to tell, but as I said earlier it's really pushing the limits. At max frequency, the opamps see a very low load impedance (100 ohms) which is not quite right. The amplitude of boost/cut is about 27dB, it can even reach 36dB in some positions, when it's typically about 15.
Is your schemo in post #42 accurate? The 100uF caps should go to ground, not virtual ground.
Many possible causes for the problem, here the issue is probably due to parasitics, like power supply sag or bad ground circulation, or not enough capacitance on virtual ground.

 

[sahib]

Maths side is extremely hairy. I wish I had the time to work out what the maximum allowed gain would be before the stability is compromised. It would be good for learning.

But Rod Elliot seems to have kept R146 , R135 and R160 to the same and for a very good reason.

I have checked the original Urei schematic and R135, R160, R125 and R214 are all kept to the same value and R146 is pretty much half of this value to limit the gain range at max, all for again a very good reason.

It may come to nothing, but I would increase R146 back to 22k again and see how the circuit behaves. If it behaves good then incrementally reduce it to work out the maximum gain allowed before things become unstable.

At some point Domingo mentions of circuit ceasing to work at certain settings and pass the signal unaffected to output. That means the transfer function is hitting a zero (again a hairy math). What I am trying to say is that there s a lot of interaction among R135, R160, R125 and R214, and now R125 and R214 being 100R you plug one hole but let the water in from the other.

As you said the problem of course may turn out to be a simple wiring error and I have talked a lot of nonsense.

 

[domingo]

Hi @sahib! I appreciate that you went through the whole thread and join the conversation.

After testing with other 'gain resistors', as you suggest, I figured that the problem starts happening only when it is about half 22K Rod Eliott's value. Also anything exactly below 11K starts failing I would say. Of course at 2K it becomes much more evident.

I've made a video to show the failure, hope it goes through. The signal is coming into the spectrum of a software parametric equalizer via a USB interface (only for visualisation), it is a real time recording of the breadboarded circuit over pink noise. An updated version of the diagram at test goes attached as well.

Is your schemo in post #42 accurate? The 100uF caps should go to ground, not virtual ground.

Yes @abbey road d enfer, I've testing with the caps to ground, as suggested by you before and shown in the new diagram. I've also simplified the bandwidth resistors to two (there were 3 before) and the caps are now 220u solid elcos.
View attachment parametric_equalizer-test1.mp4
In the video I start with max Q, moving from one extreme to the other of the fq range. Thanks to the 220u caps it doesn't oscillate in the lowest fq point anymore Then I move to min gain and min Q (widest bandwidth), where the problem is. Distortion starts, then the filter collapses and sort of inverts itself all of a sudden. So apparently the symptom is not that the unfiltered signal passes through really, but that when gain is too low (max cut) it flips to max gain (max boost) alone. I provoke the failure twice at the end.

I've read in an extensive discussion in another forum that gain can also be achieved placing a resistor to ground close to the boost/cutoff opamp. It suggests that some gain can be provided by the filter and the extra needed by the boost/cutoff differential amp (ie. see post #24). In my diagram this would be done placing a resistor between R351 and R359 to ground, I think. A value of 3.9K to GND is recommended when general resistors are 5.1K. I haven't tried this, don't understand the logic neither the value I should try. But seems like some light after the tunnel.

 

[sahib]

Hi @sahib! I appreciate that you went through the whole thread and join the conversation.

After testing with other 'gain resistors', as you suggest, I figured that the problem starts happening only when it is about half 22K Rod Eliott's value. Also anything exactly below 11K starts failing I would say. Of course at 2K it becomes much more evident.

Pleasure. It is a good learning for me too.
However, as I have mentioned before in the original Urei schematic it is about half of the other four resistors. So, the designer knew his limit.

I've read in an extensive discussion in another forum that gain can also be achieved placing a resistor to ground close to the boost/cutoff opamp. It suggests that some gain can be provided by the filter and the extra needed by the boost/cutoff differential amp (ie. see post #24). In my diagram this would be done placing a resistor between R351 and R359 to ground, I think. A value of 3.9K to GND is recommended when general resistors are 5.1K. I haven't tried this, don't understand the logic neither the value I should try. But seems like some light after the tunnel.

I have checked that link. I do not understand the logic in Increase the gain in one hand and attenuating on the other but it seems to be working.

However, the Guitar DooDaa guy has all the resistors mentioned above in equal value. You still have R354 and R366 different than the rest, not only that, extremely low as Abbey pointed out. All the examples that I know have R364, R366, R363 and R368 kept to the same value to make the maths manageable. I have looked up some references today and found only one transfer function for a text book non-inverting SVF, and it is already as hairy as a bear without the cut/boost stage thrown in.

Abbey is an expert on this and I would take his word that the design is already pushed beyond its limits. So, I think you may have to cut your losses and stick with 11k. It seems this is the best we can get.

 

[domingo]

I've been working on this again and made some findings. The circuit runs stable in prototype within a range of 50Hz to 14.5kHz (BW 0.082-2.24oct, Gain 27dB), basically full-range.

Can't credit myself, since it is directly based on @JohnRoberts and @abbey road d enfer suggestions in posts #53 and #57, who proposed to place a large resistor between the output of the summing amplifier and the -IN of the first integer opamp. The suggestion was to improve the pot tolerance at low freqs, but incidentally it allows a wider range by removing the resistor from the freq pots to ground. Instead of the 68k resistor suggested I used a 100k to increase the range even more.

The key to stability seems to be grounding the fpots straight to VGND instead of GND. Any capacitor between the fpots and GND or VGND will make the circuit oscillate (even in simulation it seems!).

The only problem I've encountered recently is scratchiness when changing frequency. @JohnRoberts warned me about this before, but wasn't clear on how to solve it. I'm using the OPA1642 now, which has very low noise current, perfect for those big 20k resistors setting bandwidth, but offset voltage is not the best. Since a coupling capacitor to GND or VGND produces oscillation, I'm planning to couple the wiper of the freq pots instead, which seems to work fine in sim. But I'm not sure if 10u is a good value, neither have I tried it yet. I thought it was wise to publish this progress and ask here before putting back my fingers on it.
Please, all comments on the overall design are more than welcome. Especially regarding the fpots wiper coupling capacitors (C4, C5) just mentioned.

Thanks for all the comments here and the knowledgeable masters for your contributions!

Domingo

 

[abbey road d enfer]

10uF and 330r gives a cut-off frequency of about 50Hz. It means there will be some action at 50Hz when frequency is set to maximum.
Can you post your .asc file?

 

[domingo]

Here is the .asc. I changed the opamp to a standard one, for some reason the analysis hangs with the external OPA164x.LIB and those extra capacitors.

https://www.dropbox.com/scl/fi/z12y...ey=s70yktop8duul25etz1fm12p2&st=pmhxkkoi&dl=0

I see some general attenuation with the 10u caps when the filter's gain is at max., but nothing happening at 50Hz particularly.

 

[abbey road d enfer]

I see some general attenuation with the 10u caps when the filter's gain is at max., but nothing happening at 50Hz particularly.

Correct. But look at the way the max boost/cut decreases at about 200Hz.

Changing C4 & C5 to 100uF seems to solve the problem.

I modified your .asc with potentiometers. This is done with a fine resolution step command. Each step represents 1/128th of rotation, 270/128 so about 2.2°.
.step param F 0 0.25 0.015625.step param F 0 0.25 0.015625
You may note that you have a very large leap of frequency at the beginning of rotation. , so actually the usable range stars at about 200.
You may want to use log pots there.

I didn't run a .noise but I would think noise may be a problem when frequency is set at maximum, due the the quite high noise gain of the integrators.

 

[domingo]

That's impressive and very useful. How did you simulate a potentiometer and what does F represent in the .step command? If you could share me a screenshot of your .asc maybe I will figure it out.

I noticed in use that low frequencies were very difficult to control, or impossible. Not good. But it was already challenging to find cheap 9mm 4-channel linear pots (I'm doing stereo). Getting log will be tough.

 

[abbey road d enfer]

That's impressive and very useful. How did you simulate a potentiometer and what does F represent in the .step command? If you could share me a screenshot of your .asc maybe I will figure it out.

See attached file. If you search the LTspice files, you have a few potentiometer models there.
F is for frequency. The wiper position of the LTspice potentiometer is initially governed by variable "w" (wiper), but you can rename it as you wish. Note that I have renamed the Boost/Cut potentiometer as "BC". You have to be careful with syntax. I initially renamed it "B/C" and it just didn't work.

I noticed in use that low frequencies were very difficult to control, or impossible. Not good.

Experience has told me that a ratio of 1:30 is the maximum you can reasonably expect. All my parametrics had range switches.

But it was already challenging to find cheap 9mm 4-channel linear pots (I'm doing stereo). Getting log will be tough.

I don't know. I've often found that log pots are more common than linear. Now I have no experience with quads.

 

[domingo]

Jewel! This is so incredibly revealing. I assumed so far that the stumbling block for a full-range SVF was opamp stability, but now I see how much resolution is lost at the fpot low extreme. Not even ninja fingers could make it usable.

This is to record ambient sounds. Maybe one stage around 60Hz-1.5kHz (1:25) and another 600-15kHz (1:25) would be a good combination.

I simulated back the design published in post #57 (schematic below). With a linear pot and 128 steps I get 1:20 as you predicted, but with a log pot 1:5. Does this mean that for covering the whole range with two stages linear is better? Or would you still recommend somehow using a log pot in this case?

Regarding the capacitors at the fpot wipers, 100uF shows almost linear gain. But I've been thinking also that, instead of those caps, it would be better to replace the opamps. I discovered for instance the OPA2145, which simulates nicely with very low DC offset.

 

[abbey road d enfer]

I simulated back the design published in post #57 (schematic below). With a linear pot and 128 steps I get 1:20 as you predicted, but with a log pot 1:5. Does this mean that for covering the whole range with two stages linear is better? Or would you still recommend somehow using a log pot in this case?

I would definitely recommend linear pots and setting the range accordingly. The reason? Linear pots have better relative tolerance than log pots. With the latter, you have the risk of having large variations of B/C within the range, even oscillation.

Regarding the capacitors at the fpot wipers, 100uF shows almost linear gain. But I've been thinking also that, instead of those caps, it would be better to replace the opamps. I discovered for instance the OPA2145, which simulates nicely with very low DC offset.

Actually I don't know why you added capacitors. Was it that the simulation showed something wrong or was it a real prototype?
With a 1:20 ratio, I never had any offset problems with NE5532's, nor scraping noise.

 

[domingo]

Actually I don't know why you added capacitors. Was it that the simulation showed something wrong or was it a real prototype?
With a 1:20 ratio, I never had any offset problems with NE5532's, nor scraping noise.

It was a real prototype on PCB and metal enclosure, the fpot scraping a lot but with my circuit shown in post #74. Some crazy 1:290 apparently and the OPA1642 (3mV voltage offset per datasheet). With the 1:25 I'm planning now it might be no problem, I hope. Or I might try something else,
Isn't the current noise of the NE5532 too high for this? Considering the high resistor values everywhere. Never tried it. The AD712 produced my favorite sounding EQ curves of what I've tried, but voltage noise specs ended up scaring me off.