Did everyone lie about all pass filters?

[jubbers]

Hello, so I am currently working on a project that requires two channels of audio to be in quadrature at every frequency. I tried using first and second order all pass filters (which are supposedly unity gain for all frequencies) but both left me with a filter response and I had the majority of the spectrum attenuated either with a 1st order low and high pass response or a band pass with 6dB/8ve roll off around Fc. The filters were always in quadrature at Fc and I’m less worried about that then the fact that the filters were clearly NOT passing all frequencies at unity gain. I not only made multiple circuits in the real world and did sweeps to test them with REW but I also made the circuits in eagle and circuit wizard and found the same results. A Hilbert transform is something I had used in the digital domain and it did what I wanted, but building that in the analog domain would require at least 31 bands and I would end up with a pretty bumpy frequency response. Anyway are all pass filters a lie? Is there any filter that anyone knows of that will pass all frequencies with unity gain and put all frequencies into quadrature?

Thanks in advance! Also I can post the images of my schematics and frequency/phase response tests if anyone wants to make sure I’m not lying or designing the circuit incorrectly.

 

[Bo Deadly]

All-pass filters are indeed real.

But bear in mind that if you have an Fc of say 1kHz, you'll see a phase shift above or below depending on component orientation. Meaning it doesn't phase shift over all frequencies. The phase shift depends on frequency. It's just the amplitude that remains constant (within bandwidth limitations of the amplifier). So, my point is, if you then mix the output of your all-pass filter with another correlated signal, you'll get a filter. So maybe you're doing something like that.

Definitely post your circuit.

 

[user 37518]

Hello, so I am currently working on a project that requires two channels of audio to be in quadrature at every frequency. I tried using first and second order all pass filters (which are supposedly unity gain for all frequencies) but both left me with a filter response and I had the majority of the spectrum attenuated either with a 1st order low and high pass response or a band pass with 6dB/8ve roll off around Fc. The filters were always in quadrature at Fc and I’m less worried about that then the fact that the filters were clearly NOT passing all frequencies at unity gain. I not only made multiple circuits in the real world and did sweeps to test them with REW but I also made the circuits in eagle and circuit wizard and found the same results. A Hilbert transform is something I had used in the digital domain and it did what I wanted, but building that in the analog domain would require at least 31 bands and I would end up with a pretty bumpy frequency response. Anyway are all pass filters a lie? Is there any filter that anyone knows of that will pass all frequencies with unity gain and put all frequencies into quadrature?

Thanks in advance! Also I can post the images of my schematics and frequency/phase response tests if anyone wants to make sure I’m not lying or designing the circuit incorrectly.

All pass filters are definitely not a lie, they are so real that I used them for my PhD and published an article using them. What you mention is something different, you need quadrature at all frequencies and an APF by itself wont do that, the way I used to do that is by running both singals through a cascade of all pass filters and the difference between the outputs should produce 90 degree difference between a certain error margin, the larger the bandwith, the more AP sections are needed. I still have the optimization script I used, if you are using the circuit I think you are uaing, I could provide you time constants for you to design your APF to achieve 90 degrees +/- some error in a defined passband.

Tell me the bandwidth and eror margin and I can post a circuit for you, do not expect a flat 90 degree line +/-2 degree variations are to be expected if the bandwidth in question is too large, Do post the circuit you are using.

 

[jubbers]

All pass filters are definitely not a lie, they are so real that I used them for my PhD and published an article using them. What you mention is something different, you need quadrature at all frequencies and an APF by itself wont do that, the way I used to do that is by running both singals through a cascade of all pass filters and the difference between the outputs should produce 90 degree difference between a certain error margin. I still have the optimization script I used, if you are using the circuit I think you are uaing, I could provide you time constants for you to design your APF to achieve 90 degrees +/- some error in a defined passband.

Tell me the bandwidth and eror margin and I can post a circuit for you, Do post the circuit you are using.

Thank you for your reply, for the passband, I was hoping to do all audible frequencies (20-20kHz). Here is the original circuit I built which is normalized for 90 degrees at 1K and the frequency (blue)/ phase (red) response.

I also switched out the caps for 0.001uF and got this response which I believe comes out to being normalized at 10K: Frequency response is purple, phase is yellow:

So you can see especially for the second circuit normalized to 10k for Fc that the phase response sits at about -90 degrees from 20Hz to 1K but the obvious issue is the filter response I am getting.

Edit: I am using a LM741 for the amp.

Thanks again for your help!
Jubbers

 

[jubbers]

All-pass filters are indeed real.

But bear in mind that if you have an Fc of say 1kHz, you'll see a phase shift above or below depending on component orientation. Meaning it doesn't phase shift over all frequencies. The phase shift depends on frequency. It's just the amplitude that remains constant (within bandwidth limitations of the amplifier). So, my point is, if you then mix the output of your all-pass filter with another correlated signal, you'll get a filter. So maybe you're doing something like that.

Definitely post your circuit.

Circuit and tests are posted, I’d love to see what you have to say about them!

Jubbers

 

[user 37518]

Thank you for your reply, for the passband, I was hoping to do all audible frequencies (20-20kHz). Here is the original circuit I built which is normalized for 90 degrees at 1K and the frequency (blue)/ phase (red) response. View attachment 97624View attachment 97622

I also switched out the caps for 0.001uF and got this response which I believe comes out to being normalized at 10K: Frequency response is purple, phase is yellow:View attachment 97623

So you can see especially for the second circuit normalized to 10k for Fc that the phase response sits at about -90 degrees from 20Hz to 1K but the obvious issue is the filter response I am getting.

Edit: I am using a LM741 for the amp.

Thanks again for your help!
Jubbers

That circuit wont work for what you want, hold on, I'll post the circuit you need, for a 20-20Khz BW you will need several sections.

 

[user 37518]

I found a Studer circuit I already had, which uses 1st order APF, you can reduce it further if you use second order sections by taking the product of their poles and zeros in a second order APF. Here it is. If this is too crude for you, give me a while for my optimization routine to compute something better.

 

[jubbers]

I found a Studer circuit I already had, which uses 1st order APF, you can reduce it further if you use second order sections by taking the product of their poles and zeros in a second order APF. Here it is. If this is too crude for you, give me a while for my optimization routine to compute something better.

Thank you very much! This is pretty much exactly what I wanted, except I don't need the channels to sum and I'm not entirely sure how that affects that circuit. I truly just need one channel at the original phase and one channel at 90 degrees which it seems like that circuit shows. It would be MUCH appreciated if you optimized the circuit and sent that! I really really appreciate this and also would love if you could tell me where I went wrong in my circuit and why I was seeing the results that I was.

Thank you again!

J

 

[user 37518]

Thank you very much! This is pretty much exactly what I wanted, except I don't need the channels to sum and I'm not entirely sure how that affects that circuit. I truly just need one channel at the original phase and one channel at 90 degrees which it seems like that circuit shows. It would be MUCH appreciated if you optimized the circuit and sent that! I really really appreciate this and also would love if you could tell me where I went wrong in my circuit and why I was seeing the results that I was.

Thank you again!

J

Yes, you don't need the summing circuit. Let me optimize the circuit, but I think you will need at least 12 opamps, unless I reduce it to second order sections and you will need 6 opamps, hold on, I'll get back to you, this takes some messing around with the optimization algorithm

 

[gyraf]

I used the above Studer circuit for Hilbert transform in my G24 - it works reasonably well (maybe within some 5 degrees off worstcase with decent-precision parts)

/Jakob E.

 

[user 37518]

Ok, this is the best I could achieve for the time being, 20Hz-20KHz, 90° +/- 2° error, here is the graph, I was striving for equi-ripple response. Compared to the Studer circuit which is 30Hz-16Khz +/- 3° error. I'll post the circuit soon

 

[neil.johnson]

It sounds like what you're after is a polyphase filter.

 

[MisterCMRR]

Back in the early 70s, a filter that maintained 90° between its two output was called a "dome filter." I made making two strings of single-pole all-pass filters. For the application, ±5° was close enough so the turnover frequencies were interleaved in roughly 3 to 1 ratios. Use more stages and smaller ratios to hold the 90° arbitrarily tight. These filters were used in radio communications for decades to make quadrature detectors that worked over octaves of frequency range. I used mine to maintain quadrature in a variable-frequency motor drive (from half-speed to double speed) for a 3M "Iso-Loop" tape machine. Using the dome filter, plus a high-pass filter to increase drive voltage linearly with frequency as well, two audio power amps driving the motor windings directly gave it full torque over the 4:1 speed range.

Such a filter is at the heart of the attached circuit I found in a quick web search. For what you're doing, drive the lines marked "L" and "R" together. Output of the filter is on lines marked "I" and "Q" (traditional for In-phase and Quadrature).

 

[sdimond]

Hello, so I am currently working on a project that requires two channels of audio to be in quadrature at every frequency.

Are you trying to build a frequency shifter? There is a considerable published body of work about the topic of quadrature phase shifters. Starting with: Wideband Phase Shift Networks by R. B. Dome in Electronics magazine in 1946. If you want I can give you additional references

 

[MisterCMRR]

Thanks sdimond for the reference!! Now I understand why they're called "Dome filters" ... my mystery of the day solved!

 

[MisterCMRR]

And here's a copy of the article from Dec 1946 issue of "Electronics" magazine (I was only 2 at the time!)

 

[abbey road d enfer]

I truly just need one channel at the original phase and one channel at 90 degrees

Which you won't get. Wide-range phase-shifters rely on phase difference between two APF's, so no signal is left untouched. I won't start a debate about audibility of APF's but you must be aware of that when you use these signals. In particular mixing either of these signals with the original will result in strange response.
This type of circuits has been used in the Vox AC30 vibrato, in the Surrey Electronics Spectrum Shifter, and of course in the aforementioned Studer circuit that was supposed to solve a non-existent problem.
A tangential use in compressors side-chain exploits the trigonometric property sin²+cos²=1 for providing automatic attack time. I believe the only practical implementation is done by gyraf IIRC.

 

[zamproject]

Hello abbey

Studer circuit that was supposed to solve a non-existent problem

Do I miss something about this one ? To me it seem legit for mono summing compensation

 

[abbey road d enfer]

Hello abbey

Do I miss something about this one ? To me it seem legit for mono summing compensation

FROM THE DESCRIPTION:
"Doubling of equally-phased signal components as well as canceling of opposite-phased components is thus avoided."

• Doubling of equally-phased signal components ... is thus avoided. This is only a matter of normalizing levels, which is needed anyway, depending if the mono signal is listened to on one or two speakers, and what speaker. Quite often mono listening is done via a separate speaker.
• canceling of opposite-phased components is thus avoided. Yes, but what about quadrature-phased components? They end up being cancelled. I tend to think that in a good mix, there are more quadrature signals than out-of-phase

I have much respect for the german broadcast designers, but sometimes they have made mistakes, like the constant-power pan-pot.

 

[zamproject]

Yes, but what about quadrature-phased components? They end up being cancelled. I tend to think that in a good mix, there are more quadrature signals than out-of-phase

exactly what I miss ! tanks for the light

 

[Holger]

I have much respect for the german broadcast designers, but sometimes they have made mistakes, like the constant-power pan-pot.

Interesting. Can you provide additional information on that or point to some links about that topic?

 

[abbey road d enfer]

Interesting. Can you provide additional information on that or point to some links about that topic?

The constant-power pan-pot concept is based on constant global power, so when the PP is at center, each speaker receives half the power compared to when the PP is full right or full left.
But actually audition is not based on power, it's based on sound pressure.
Sound pressure is not a vector, it's an algebraic scalar (thanks user 37518) value. So the combination of two identical pressures is twice the pressure, or a 6dB increase.
So, a centered PP with result in a 3dB boost when in the centered position, which is not correct.
The solution would be to have a constant-voltage PP, where the level at center is 6dB down, and the result would be mostly correct. that results in 1/4th the power in each speaker, for a total of 1/2.
One may wonder how it is possible to produce the same SPL with half the total power. Seems to defy a basic law of physics, energy conservation.
The answer is that the directivity pattern of two separate sources is not the same as a single source.
The conservation of SPL will be verified in the median plane only, and there will be zones with much less SPL, due to cancellation.
Validity of the concept depends on how the pressures from separate sources are correlated, or how accurately two sources combine and result in doubling.
Measurements have been conducted in many studios and auditoria that show that the correlation is very good in most studios, with an average of about 5.5dB, which clearly shows that the constant-power PP is inadequate.
Correlation is not so good in many venues, with a an average of about 4.5dB instead of 6. It's predictble, since the speakers are usually much more separated. That's the reason why many live desk manufacturers opted for PP with -4.5 dB in the center position.