True Tape Thru Zero F****ing - perfect?

[puretube]

25315


did I go too far, here?
:wink:    :idea:  :!:

[Vsat]

Depends whether you sum or difference  the two signals... addition gives 6 dB boost at the zero-point. Stm's circuit gave excellent cancellation in difference mode.
Mike

[puretube]

talking bout differential mode here...

is total cancellation more desirable than accelerated zipping through outphasing harmonics?

[Vsat]

Not necessarily more desirable - just different.... certainly don't want to stop at the zero-point for very long in difference mode. Lots of dispersion so thru-zero does not occur simultaneously for all frequencies, but rather progressively moves across the spectrum can be great sounding. Ideal situation would be to have both options available.
Mike

[puretube]

aaaahhhh,
that last sentence is what i hoped to one day hear from a person like You!


the 2 pics in the other thread include both options continuously alternating/morphing.


gotta shut up, now :oops:

[stm]

Hi, in my opinion a "perfect null" is less musical than a partial but intense null.  Just think of other non-ideal things we like:

1) Clean channel "not so clean" (something like 2 to 5% THD)

2) Analog delay with restricted bandwidth instead of 20 kHz crystal clear bandwidth.

3) Some compression/limiting instead of infinite dynamic range.

Now regarding to the analog delay line, Mike built the second design with 6 kHz bandwidth.  With the latest design I posted, you can have 8.5 kHz BW at 0.5 msec delay.  Furthermore, I already devised a simple method for fine-tuning the whole thing for maximum amplitude flatness with only three trimpots.  I'm pretty sure the above may allow you to get even better cancellation.

It's up to you to decide if you want that much or not. The solution is very simple, instead of subtracting both signals with the very same gain, just attenuate one of them slightly until you have the desired amount of cancellation (or feedthrough).  Anyway, BBD's don't have perfect 0 dB amplitude transfer, so anyway a trimpot is in order.

:wink:

[Vsat]

Also, some nice "Eventide-style" stereo tricks should be possible if use is made of a tap between the two series-connected  250 uS sections in the network...
Mike

[stm]

Interesting point!

Don't forget an SAD1024 has an intermediate tap also. What if:

1) 1st BBD tap +/- 2nd Delay tap -> left channel
2) 2nd BBD tap +/- (or even -/+ !) 1st Delay Tap -> right channel

I think that would be "maximum stereo brutality"  :twisted:

[Vsat]

stm,
SAD1024 would be compact way to do it! Or just use a 2nd MN3009... Also MN3001 and MN3010 very hard to find nowadays.

puretube - BTW nice looking PCB - when do we get to hear it in action?
Regards, Mike

[Chico]

Puretube:

I see what you are saying about Fc changing with tape speed in the case of a tape flanger. Could controlled variable filtering be used to better approximate what is going on in tape flanging?  

Didn't MXR have a variable frequency antialiasing filter using some form of PWM on a flanger or delay?

I was thinking of a simple approach using an off the shelf switched capacitor filter chip, e.g., by Maxim.  The 100:1 chips should stay out of the audio range if properly configured, I would suspect.  The idea would be to have two parallel BBD lines, each tuned to a close, but different base fc, e.g., using a 4046 PLL or similar chip.   Each switched capacitor filter fc would be further modulated by an LFO that corresponds  (or does not correspond) to the delay of the chip driving the CV input of the PLL.  

The LFOs to the switched capacitor chips could even be used to exaggurate the differences in delay.  And one could even use different orders, filter topologies etc. to change the phase characteristics over time as the fcs are modulated so that the delay lines "look" inherently different even at the same delay time.

Any thoughts on this approach?  Obviously one would have to be careful to balance the changes in the filter to the clock of the delay line, and I can imaging a parade of horribles introducing yet more HF clock signals.

Best regards

Tom

btw, Puretube, that circuit looks pretty cool.  Not only are you several steps ahead in the game, but you have some tricks up your sleeve that we have yet to get into here.   Morphing?  that is forward thinking.

[puretube]

...when do we get to hear it in action?
Regards, Mike

it was demoed earlier this year @ Frankfurt MusikMesse:

(the rightmost "censored" box...)

to a few critical listeners:


.....

ps: concerning "taps": that thingy got 3 stereo output jacks...

[puretube]

...some tricks up your sleeve ...

well, only some (published) considerations of "good ole guys"
(whose names I repeatedly dropped here...),
hopefully well-understood, implemented in a different way,
mixed, and packed into one box...

[stm]

as opposed to the following stm-post, saying: "a
time-delay is a time-delay", which is not true!
(can be seen in the phase vs. frequency plot that i asked for... :
a tape-delay would only show a slight phase-variation over the range,
while the AP delay moves thru 2 complete circles).

I took a deep thought before posting this; initially I was inclined towards the easy road of stepping aside and don't bother arguing, but I think it is neither good for the people nor responsible to leave a comment that can be misleading. Please notice that my intention is to look for the truth and not to cause controversy.


I think there has been a misunderstanding with phase plots and their interpretation, which I will try to clarify here.  I apologize for the use of mathematics and Laplace, but is the only scientific way I found to demonstrate my point.


1) I sustain that a "delay is a delay" no matter if it is generated by an all-pass filter, and ideal delay, or a tape playback unit, provided that:

  a. Audio signals lie withing the operating bandwidth of these devices.

  b. Amplitude response is relatively flat for these devices within said bandwidth.


2) An all-pass filter has the following characteristic curves:

  a. Amplitude response: flat within the band of interest and usually further.

  b. Group delay: flat up to a certain frequency and then starts falling down to zero.

  c. Phase response: linear from zero up to a certain frequency, and then starts leveling at a final value equal to -90º x N, where N is the filter
order.  To see the linear response you must use a linear scale for the phase and frequency axes.


3) An ideal or "pure delay" device has the following characteristics (which you can plot in a circuit simulator using an "ideal delay" device):

  a. Amplitude response: flat from zero to infinite frequency.

  b. Group delay: flat from zero to infinite frequency, equal to the time delay specified for the "pure delay" device.

  c. Phase response: linear from zero up to infinite frequency. A straight line whose slope depends on time delay specified for the "pure delay" device. It has no limit, reaching (-infinite) phase at infinite frequency.

  d. This will be used later, but I prefer to introduce it here.  The time delay operation is represented in the Laplace domain as: exp(-s*D), where s=i*2*PI*f is the laplace variable, and i = sqrt(-1).


4) A tape playback device is supposed to have flat amplitude response, flat delay response, and nearly flat phase response within its operating
bandwidth.

  a. The above statement is true if you consider the tape playback unit as a blackbox and measure its frequency response curves disregarding the time delay on its output.  Then, one could say that its phase response is very different from a time delay implemented using an all-pass filter, thus
concluding that both delays are "of a different kind".  Wrong!

  b. We are talking here about a DELAYED signal coming out of this unit, and as such, this has to be taken into account in the transfer function as well, otherwise we are comparing different things and the statement is invalid.

  c. Let's assume the tape recorder has a nearly flat freqency response represented by a transfer function called F(s).  Now, to consider that the
output has a time delay of D seconds, we must include the time delay operator presented in 3)d. The resultant transfer function is simply the
multiplication of both: exp(-s*D)*F(S).

  d. If you take the curves of the above function, the exp(-s*D) term alone provides a linear phase to frequency relationship which will be very close to the phase of an all-pass filter that introduces a similar delay of D seconds, up to the point where the all-pass filter no longer behaves as a pure time delay. In summary, both phase responses are very close to identical within the band of interest. Q.E.D.


5) Other common "trick" that a phase plot can play you is:

  a. If you see a phase plot of an OpAmp inverter, it will show you -180º. We now an inverter introduces nearly zero time delay, so we cannot simply calculate the delay as half the period of the signal.

  b. Now, if you have two first order all-pass stages in cascade (like the ones implemented with FETs or photocells on phasers), each producing -90º phase shift at the frequency under study, again you have a total of -180º phase shift, but this time you have indeed a time delay!

  c. The moral of this is that you have to be very careful on how you use and interpret phase information.


Best regards,

STM

[Mark Hammer]

It's thorough responses like that which make this a great thread.

The best opinions are those expressed so articulately that you can learn something from them if if they don't change your mind.

[puretube]

gone info...

[stm]

As stated in 2.c, to see a linear response you need to plot with linear phase and frequency axis.  The plot you are looking at has log frequency axis (it was my fault I didn't use linear axis).

Here I am including a simulation of my latest 0.6 msec circuit.  The blue phase curve is flat up to 8 kHz, where the allpass network stops behaving as a pure delay.  For comparison, I added the phase plot of an ideal delay element (included in the simulator's library), which is in red.  The latter curve extends to infinity.  Now I'm using linear scales.



Also, by definition the group delay is defined as the derivative of the phase function 'Phi' with respect to the angular frequency 'W':

GD = dPhi(W) / dW

In the linear region we can just approximate dPHI(W) and dW by the slope of the line (which passes through zero), so:

GD = Phi(W) / W

We'll take the measurements at 8 kHz, so Phi(8000 Hz)=1700 degrees. They must be expressed in radians to be used in the above formula, so, Phi(8000 Hz) = 1700 / 57.3 = 30 rad.

Also, the angular frequency corresponding to 8 kHz is 2*PI*8000 = 50266 rad/sec.

Then, GD = 30 / 50266 = 597 usec, pretty close to 0.6 msec.

[Vsat]

Surprised this paper hasn't been mentioned yet:

"Flanging and Phasers" W.M. Hartmann, Journal of the Audio Engineering Society, June 1978, Vol.26, No.6,pp.439-443

Regards, Mike (tried to attach a "winkie" but no java on this computer)

[stm]

Mike, do you have an address to see that paper?  

[puretube]

gone info...

[Vsat]

Summary: Hartmann concludes that phasing becomes equivalent to time-delay flanging in the limit when an infinite number of allpass stages are used, each producing an equal (but infinitesimally small) delay.... Hartmann investigated several optimization schemes whereby a finite number of  stages were tuned differently in an effort to produce notch/peak positions identical to flanging... using identical stages gave the best (still far from ideal) results. A big caveat however: this paper talks only about FIRST-ORDER allpass sections, Sebastian uses
SECOND-ORDER which gives an additional degree of freedom to optimize within a given passband.
Regards, Mike