MN3207 + MN3102. How to test

[DrAlx]

Adding two identical signals which have only different delays gives a comb filter effect with peaks separated by 1/(relative delay).
So realtive delay of 1ms will give peaks separated by 1kHz (with peaks and troughs all the way from DC to infinity in theory). That's why peaks move when you change relative delay.

The peaks are getting weaker at higher frequencies because you have quite heavy filtering on the BBD outputs.
That may be deliberate in this circuit.

If you changed the 33n to 3n3 you would see stronger ripples at higher frequencies.

[temol]

DrAlx - I totally forgot to mention that now I have only one DL, so no interaction between DLs. Sorry for this..
Schematic is also different.
I doubt it's about filtering after BBD. I bypassed the filtering (signal from BBD directly to summing opamp) and still the same effect. In fact it does not matter if it's previous or current schematic. Same story.

T.

[DrAlx]

You are still mixing 2 signals with different relative delays and so will get comb filtering.
There is the "clean" signal that avoids the BBD and so has zero delay, and the delayed signal out of the BBD. You get a comb filter response with peaks and troughs when you sum these with U3A.

If the delayed and clean signals were of equal strength before being summed, then the comb notches would be very deep and give almost complete cancellation at some frequencies.  The fact that you don't see deep notches means that the signals are not of equal strength before they are mixed.  Also since the notches are practically non-existent at higher frequencies, this means the signals are even more mismatched at high frequencies. You can check for yourself...

Measure the frequency responses separately at the outputs of U2B (the clean signal) and U3B (the delayed signal). i.e. look at things before they are added.
You will probably see outputs of different strengths, and these will become more mismatched with increasing frequency.

EDIT.  The setting of TR1 has an effect too.  I'm assuming it's set to zero ohms so the two outputs are being mixed equally.


2nd EDIT.  And I am wondering why R35 is connected to +ve supply instead of ground??   For MN3207 I would expect the resistor on the BDD outputs to go to ground.  Has this circuit been adapted from an MN3007 design?



Looking at the input to the BBD, you have a voltage divider from the 1k and the 3k, so signal is getting scaled down by a factor of 4 at input.  Not a good idea unless the input signal would otherwise cause the BBD to clip.  Can't tell if that's the case here.
Also I see there is no BBD bias control.  The assumption is that 5V is correct bias.  Not a good idea.
5V may be close to OK bias in this case, but if bias is set poorly then you have to reduce input signal strength to avoid the BBD clipping.

[temol]

There is an error on the schematic, in my layout R35 goes to the ground. Before mixing signals are not equal in strength. There is a signal meter in RMAA - when I set clean signal to -2dB then delayed signal, with no attenuation is -11dB. But as soon as I start changing signal mixing ratios, increasing level of the delayed signal, freq response is getting too wavy in LF region (see the attached screenshot).




ps. I do not know origins of the circuit. It's not mine design.

T.

[ElectricDruid]

It's not just a question of filtering after the delay line. There's also the anti-alias filtering ahead of the delay line, which is important if you don't want the signal mangled. And if there's no highs going in, there's no highs coming out, even if you do bypass the output filter. Well, except for the highs produced by the sampling process itself, which we don't really want.
Where's the input filter set? What's the BBD clock frequency? How much high frequency content can we reasonably expect?

T.

[DrAlx]

Looking at the short animation you can see the comb notches move left as you increase the BBD delay.  The notch spacing is uniform in frequency but since you have a log horizontal axis the notches bunch up as you look right.  That's in agreement with theory.

Now I noticed that the amplitude of the comb notches appears to be related to how bunched-up they are horizontally in the animation, but that is not backed up by theory.  In other words, if you look at the 1kHz area of the plot and vary the delay then the local ripple width should change (and it does) but the local ripple amplitude should not change (but it does!!!)

I am wondering if this is an artefact related to how things are measured (which involves sampling), and then processed to give the frequency plot?

[temol]

I attach schematic of the DL setup I have on a breadboard now.



With "dimension" pot set to zero (R3 only), there is a 225 kHz on pin 2 of the MN3102. With pot set to 35k (68k total) - 116 kHz. Pot set to max ~ 38 kHz.
I have no idea if this might be related to measurement and processing of the signal. I am using RMAA software and onboard soundcard.

T.

[DrAlx]

225 kHz on pin 2 gives notches spaced by 225kHz/512 = about 440Hz at the final circuit output.

I know it's not the same circuit but check out this image of a comb filter effect for a flanger, where the notches are spaced by about 400Hz.

http://www.metzgerralf.de/elekt/stomp/mistress/images/measure/comb_dem2.gif

The bunching up of notches to the right is just from using a log scale for frequency, but notice that the notch depth also decreases with increasing frequency.  It's not as extreme a decrease as in your graphs but it's there.

Sticking with this gif image, I can't tell how much of that decrease in notch depth is real (i.e. due to a LPF somewhere affecting things) or if some of it is due to a sampling/processing issue with the software used to make the image.  I suspect there is a sampling issue.  Why? Imagine drawing a line connecting the peaks in that image.  It would be quite smooth.  Now imagine a line connecting the notch bottoms in that image.  It would not be smooth but have many bumps.  If a LPF was affecting things I would expect it to affect the top and bottom envelopes in a similar way, or at least affect them both smoothly.  That's not the case, so I think there is definitely a problem with insufficent sampling in the frequency domain in producing those images.

e.g.  up to about 2kHz in the picture, it's pretty easy to see the notch shape because the pixels are relatively small compared to the notch width, but by the time you are at 10 kHz it gets harder to hit the extreme dip of each notch.  A pixel is more likely to hit one of the steep side-walls of the notch instead of its very bottom, and so the drawing will show the notch to be less deep than it really is. 

Now you have not produced your graphs using the same software as that gif, but the same sort of thing could be at play, because in theory you should expect variation in notch width as you change the delay (as you have seen) but notch depth in a region (say near 1kHz) should not change as you vary notch width.

[temol]

Now the question arises (at least for me) - why the notch amplitude decreases with frequency increase, and why it dissapears completely from HF range. Maybe it's obvious for someone that's oriented in physics of the sound, etc. I'm going to check other software with ability to take frequency response measurements, just to rule out software restrictions.

T.

[DrAlx]

...why the notch amplitude decreases with frequency increase

I don't think that's the real issue.   Your animation shows that even if you restrict yourself to looking at a small range of audio frequencies close to 1 kHz, then you have a graph that shows notch depth decreasing as you increase delay. Here's a simple thought experiment to highlight why that is not realistic.

Take the simple case when you have an audio signal at exactly 1 kHz (i.e sine wave) and a BBD that provides pure delay with no attentuation.
For 1kHz audio the signal period is 1ms.  If the BBD is set to give a delay of 0.5ms then when you add the clean and delayed signal you will get complete cancellation because the BBD has shifted the wave by half a wavelength.  So on your graph you would have a point at -infinity dB (i.e. a perfect notch).  For a longer BBD delay of 1ms the exact opposite happens because the two sine waves are in phase so the signal is doubled when they are summed, so you get a 3dB gain.
Now consider a much longer delay of 10.5 ms.  What happens then? The BBD shifts the audio by an odd number of half wavelengths, so you get perfect cancellation again.
So you see the act of increasing BBD delay over a large range causes the 1kHz output signal to repeatedly cycle between two extremes (perfect addition and perfect cancellation).  Now if the BBD gain is less than 1 (as in your circuit) then the you wont get perfect cancellation or addition, but the basic principle still applies.  i.e. the output signal will repeatedly cycle between a min and max level.  The min and max levels themselves do not depend on the BBD delay.

Your animation does not display that behaviour.  You can clearly see the ripples near 1kHz getting smaller with increasing delay.  Rather than rely on software to try and get a graph, I would simply drive a pure 1kHz tone into the circuit and try to measure the output amplitude of the 1kHz sine wave as you vary the BBD delay.  Ideally you should use a scope to do this.
You will see the output wave amplitude vary between two levels.  it will be hard to determine those levels when the delay is large because the amplitude will change very quickly even for small changes in BBD delay pot setting.  The difficulty you have in trying to manually measure the min and max wave amplitude is the same sort of problem the software has in trying to locate the notch minima.

[allesz]

The last schem is quite similar to the echopathetic... It should sound "decent" at least.
I settled to a 100nf delay time cap and I get no clock noise.

Of course is hard to expect hi-fi from such simple circuita.

[temol]

I took a Visual Analyser (software oscilloscope) and compared two signals, clean and delayed. It's a zoomed view, so you can see some unevenness to the sine waves. As you can expect, while rotating "dimension" potentiometer, delayed sine wave starts to "shift". 


But then I took a reading from cabsim output using Room EQ Wizard.


And now we have a different look of  the frequency response graph. I'd say, proper one. In REW there is one setting responsible for the amount of the detail visible to the graph - smoothing. You can compare the graphs with/without smoothing and with two different levels of smoothing.

1/6 smoothing vs 1/48 smoothing.


1/6 smothing vs no smoothing


I also attach spectrum of the signal  (from  Audacity).


T.

[DrAlx]

So old software was smoothing the curve and so causing narrow ripples to be flattened out. Explains the odd animation effect nicely.

[temol]

Exactly. Thank you all for your help.

T.