Vari-Q Bandpass Filter With Constant Gain

abbey road d enfer said:

Is it the phase response of the speakers, or the phase response of the speakers + cross-overs? It is important to figure out what causes the phase-shift. Very often, cascaded 1st-order APF are preferred to higher-order, because they are unconditinally stable and they are very easy to align, using a single pot.

Click to expand...

I always thought the mid band was the most problematic area in filter designs when it comes to phase issues due to the amount of components (and in DSP , code but I think it shouldn't be a problem with todays processor clock rates)

Then again, how should I know..

mjrippe said:

Telling someone "just google it" is not teaching.

Click to expand...

But Abbey is not  a teacher, and neither are PRR, etc. I have often wondered why in a world where virtually nothing is for free, some people take their time to help so much. I don´t see any obligation for them to do so. People have  helped me a lot and showed a lot of patience as long as I was showing to do a little homework.

Btw. look at other forums, this one is very well moderated.

It´s a little problem if people cannot tune in to the tone and topic of this forum. If someone posts *I want to build an CMOS oscillator but don´t know how to solder*! Or like in that recent "I want help" thread (well that may be a language barrier thing, but I understand the way Khron has replied to that)EDIT: YES, IT WAS. Noone here is sitting and waiting in a 5cent psychology booth ;D

Ok back on topic

Look, I know moderating is a thankless job.  I also know that Abbey shares a HUGE amount of knowledge here.  I am just saying politeness will make this a better place for everyone.  Sorry for dragging the issue out.  As L' Andratté said, back on topic.

The phase shift only occurs on the left side midbass driver, and it's a function of many things IMO. It occurs outside of the crossover region, in the passband of the driver. The biggest part of the problem is that it's in a car door.

The driver is about 75 degrees off axis from my ear, so as frequency increases and the driver becomes more directional, the reflected sound (path: driver's door mounted speaker => passenger side window => drivers side window => ear) becomes predominant. My leg is also smack dab in front of the driver, so it absorbs and diffuses and generally makes a mess of things. The driver's side door glass and windshield also contribute something, it's probably too complicated to really define as the contours of the car are so complex. The result though is a phase shift of approximately 180 degrees occurring over a band from about 100 hz to 200 hz:


So far though I haven't found a way to shift the phase of the delayed signals back in time. Perhaps a DeLorean? Instead, I took the approach of altering the phase curve of the right side driver via an all-pass filter in order to match the phase curve of the left side.


Then added in an all-pass to the left side to pull the phase curve down above 400 hz. With that approach I could get the phase curves of each side to line up relatively well from 20 to 600 hz:


This was all done using MFBP second order all-pass circuits, and powered by a cute little premade dual voltage supply. I made the measurements with REW, then simulated the all-pass filters in REW, and then built and tested the filters.


The end result was very good. Still not perfect, but things like bass guitar lines were much more consistent throughout the frequency range and more centered on the stage.

The problems though:
1) My measurement method of measuring at one ear then the other probably had some error in it.
2) I found that the phase curve of the drivers tend to change with volume. It seems to follow a few defined paths though.

So my plan now is to play with an all-pass that can be adjusted while seated in the drivers seat and see what else I can learn from that.

mojozoom said:

So far though I haven't found a way to shift the phase of the delayed signals back in time. Perhaps a DeLorean?

Click to expand...

Probably; it's impossible to create negative propagation time. Every time alignment relies in delaying the earlier path.

Instead, I took the approach of altering the phase curve of the right side driver via an all-pass filter in order to match the phase curve of the left side.

Click to expand...

It seems to be working not too bad.

The end result was very good. Still not perfect, but things like bass guitar lines were much more consistent throughout the frequency range and more centered on the stage.

The problems though:
1) My measurement method of measuring at one ear then the other probably had some error in it.

Click to expand...

For signals in the 200-600 Hz range, that should not be a big issue, since the 1/4 wavelength is about the same as the interaural distance.

2) I found that the phase curve of the drivers tend to change with volume. It seems to follow a few defined paths though.

Click to expand...

The only explanation I can think of is that a panel resonates non-linearly.

Good afternoon equalizer experts! I've spent some time playing with this circuit now and have some questions that are holding up my board layout, so I'm needing a push in the right direction again.

I was looking at the way Rane implemented this in their 1991 SP15 model, and came up with a few questions. Here's the Rane implementation of the state variable bandpass in the SP15:



1) They applied a 100uF electrolytic between the bandwidth adjusting pot and ground. When I model that it adds some noise to the amplitude plot at the high Q points. I can minimize the discontinuity by bumping that cap up to say 470uF, but that ends up being the largest device on my board.  Can anyone advise the need for the cap in that location? I wouldn't expect Rane to put it there if they didn't see a need for it…

2) Although the Freq is controlled by a 100k pot, they seem to consistently apply a minimum 5k-6k resistor before each integrator to fix the max Freq you can adjust to. On some other circuits I've seen them go as low as 2k, but no lower than that. Is there minimum resistance to strive for there that helps keep the op amp stable, or something similar? For my application I was looking at using 100 ohm there, not 2k-6k.

3) They use a 100k pot on the Freq adjustment. Wouldn't there be benefits in noise reduction in working that down to a 50K or maybe even 10k?

Thanks!

mojozoom said:

Good afternoon equalizer experts! I've spent some time playing with this circuit now and have some questions that are holding up my board layout, so I'm needing a push in the right direction again.

I was looking at the way Rane implemented this in their 1991 SP15 model, and came up with a few questions. Here's the Rane implementation of the state variable bandpass in the SP15:

1) They applied a 100uF electrolytic between the bandwidth adjusting pot and ground. When I model that it adds some noise to the amplitude plot at the high Q points. I can minimize the discontinuity by bumping that cap up to say 470uF, but that ends up being the largest device on my board.  Can anyone advise the need for the cap in that location? I wouldn't expect Rane to put it there if they didn't see a need for it…

Click to expand...

The DC blocking capacitor is to prevent control noise from changing DC gain. If you don't care about DC wiper noise go ahead and ground it, but confirm that DC operating point doesn't compromise headroom.

2) Although the Freq is controlled by a 100k pot, they seem to consistently apply a minimum 5k-6k resistor before each integrator to fix the max Freq you can adjust to. On some other circuits I've seen them go as low as 2k, but no lower than that. Is there minimum resistance to strive for there that helps keep the op amp stable, or something similar? For my application I was looking at using 100 ohm there, not 2k-6k.

Click to expand...

no minimum (ignoring current drive capability). I am not a fan of rheostat connected frequency pots in SVF and have discussed alternate potentiometer approach numerous times in other threads.

3) They use a 100k pot on the Freq adjustment. Wouldn't there be benefits in noise reduction in working that down to a 50K or maybe even 10k?

Thanks!

Click to expand...

Noise will likely be dominated by op amps ein and topology... there are alternate connections to vary the Q/bandwidth that deliver different noise gain.  For example for wide bandwidth spreading the two SVF poles further apart will deliver lower noise gain than bandwidth pot to ground.

There are many ways to skin the same cat with SVF but perhaps get a handle on how low you need the noise floor to be, before killing brain cells (or BOM cost)  in pursuit of "lower must be better".

JR

Thank for taking the time to help me John. Obviously I'm no pro at this stuff so it's really nice to have fellows like yourself that are willing to give me some tips.

The source is digital, and after that is a separate DSP claiming THD+N of <0.01% @ 1 kHz, and SN of >105 dBA, so I'd like to be in the realm of that. I really don't know what constitutes the limits of audible noise levels, so I can't really say what to shoot for here.

I reworked the circuit based on the posts you referenced and also some schematics on Rod Elliott's site. The results are pretty good at this point, with just some minor ripple at extreme Q settings. Currently the Fo is adjustable from 52 Hz to 1.2 kHz, and Q from 0.6 to 7.4. The noise analysis in LTSpice is showing around 0.09% THD, if that can be trusted.

I've got 100uF caps on the ground path for the frequency control pots, but currently not on the Q control at that was causing a pretty big ripple in the results.  I also shifted most of the resistors down to 10k as this isn't really a standalone item but will be receiving signal from an INA1650 so I figure input impedance shouldn't be as critical.

mojozoom said:

I've got 100uF caps on the ground path for the frequency control pots,

Click to expand...

  I'm not sure it does actually solve the issue of scratching pots since the opamp offset is still circulating in the input branch of the following opamp; I would think best to put the cap between the oamp output and the resistor that drives the pot. Then the only DC current would be the input bias current in the inverting output. It may be necessary to also DC-isolate this inverting input, but it wouldn't take a big cap. These caps may produce "ripple", though...

but currently not on the Q control at that was causing a pretty big ripple in the results. 

Click to expand...

Really? How "big"? Can you describe?

Guy's I'm sorry - I didn't tell you that the ripple I'm seeing occurs when I finish out the circuit by including the summing amplifier. Here's what the amplitude plot looks like without the capacitor in the Q adjustment ground path, and then with the capacitor.

Without:


With:


The ripple is highest at the extremes of Q.

mojozoom said:

Guy's I'm sorry - I didn't tell you that the ripple I'm seeing occurs when I finish out the circuit by including the summing amplifier. Here's what the amplitude plot looks like without the capacitor in the Q adjustment ground path, and then with the capacitor.

Without:

With:

The ripple is highest at the extremes of Q.

Click to expand...

OK, that shows that in order to minimize these, the capacitor value should be 10x. But, as I wrote before, that would not solve the issue of scratching frequency pots.

Ok I remodeled it with a 1000uF cap in the Q adjustment and that brought the ripple way down. Unfortunately that's a really big cap though, so I'm going to go with a 330uF cap as it gives a ripple of about the same magnitude that I'm stuck on toward the top of the frequency adjustment range.



For the pots I've read that conductive plastic can help minimize the scratchiness so I'll give that a shot here. It's about $1 more per pot and I only have 4 in the circuit so not too big a deal.

Thanks again for all your input and guidance everyone. I'm working on the board now...

I believe Abbey's point was that the scratchiness was due to the pot still affecting the DC gain of the circuit.

Good luck

JR

My parts are in now, but I'm still waiting for the board. This will be my first all SMD build.

In order to avoid pots, it sounds like I'd need to use digpots, MDACs, or VCAs. If I was to implement VCAs, would I only need to apply them in front of the two integrators and leave the bandwidth adjustment pot as is? Or is there value in changing the bandwidth to VCA control as well?

Thanks!

I would lean toward digital pots and depending on the Q range you want you might get away with only two, one on each integrator stage.

If you are feeding digital into a DSP, could you perform the EQ inside the DSP?

JR

Unfortunately the DSP is a manufactured unit with a limited interface and no additional development on the interface seems to be planned. I don't believe I can get access to the programming of the thing as they have a way to lock it with an access key. Maybe someone on these forums has experience with that type of thing.

So it functions as an 8x32 band parametric EQ with crossovers and delay, which is pretty sufficient, but without an allpass.

I was initially considering using digipots with i2C for control via an Arduino and bluetooth interface, in order to allow sitting in the listening position while making adjustments (via smart phone). But some of the digipot literature seemed to indicate that there could be zipper noise with them, and I didn't think it would resolve the issue of the dc pot noise. Or maybe that's not a concern when there isn't a physical wiper in the pot?

mojozoom said:

But some of the digipot literature seemed to indicate that there could be zipper noise with them, and I didn't think it would resolve the issue of the dc pot noise.

Click to expand...

Wiper (DC) noise and zipper noise are two different (though related) issues. Discretization od the pot's variation creates artifacts when DC circulates (this can be obviated with standard capacitive coupling, just like wiper noise) and also when signal is present. In the latter case, is it a big issue? Unless you want to sweep the EQ for creative effects, the effect would not be much different than sweeping a pot, i.e. something that you typically don't want to do while there is signal.

Maybe find some different literature... Modern high performance DPOTS with adequate resolution can also incorporate zero crossing coordination... If DPOT code changes only occur during signal zero crossings there are no clicks.

JR

Thanks John - I dug around a bit and found one good candidate from Maxim that does just that, the DS1882. It's the only one I've found so far with log taper, NV memory, and +/- 7V analog supply. All of the others I'm finding have a 5V supply max.

> with log taper

Note that, within limits, a "log taper" can be done with a linear pot by skewing the number put in the digital side.

Given a 256-step linear, these numbers give uniform "log" response:

256 = 0dB
128 = -6dB
64 = -12dB
32 = -18dB
16 = -24dB
8 = -30dB
4 = -36dB
2 = -42dB
1 = -48dB
0 = -infinity

It gets a bit coarse below -20dB but you have <0.5dB resolution down to -20dB. Which is a likely range for an EQ. If you truly need a "forty-some dB dip", you probably do not care if it is -42 or -45dB.