I think that we are losing track of what 'might' be important. The retention of the 'shape' of a pulse might give us some useful data, but it has been shown that some phase shifting cannot be easily detected, yet other phase shifting or (tone inversion) can be readily detected. There must be something more than just the overall 'look' of the waveform that is the most important and somehow our ears detect it.
Related to this, might be the information measured by the 'Hirata test' that uses several versions of asymmetrical waveforms and the results appear to track what the human ear hears. For example, the test shows that Marantz tube amps measure well, and the Crown DC300 power amp measures very poorly. Why this is so, is beyond me at the moment. Dr. Hirata was ahead of his time, I'm afraid, and after a few years, has appeared to drop his efforts in this direction. I was impressed, however.
Related to this, might be the information measured by the 'Hirata test' that uses several versions of asymmetrical waveforms and the results appear to track what the human ear hears. For example, the test shows that Marantz tube amps measure well, and the Crown DC300 power amp measures very poorly. Why this is so, is beyond me at the moment. Dr. Hirata was ahead of his time, I'm afraid, and after a few years, has appeared to drop his efforts in this direction. I was impressed, however.
And you think his different pulse responses won't show different harmonic spectra?
Pulses with THE SAME harmonic spectra can have different look, and the role of linear distortions (phase shifts) is easily visible just by the change of pulse shape. In theory, with a math model of an amp, the nonlinear and linear distortions will be related somehow, dependent on the model, but with a real amp, I believe, many surprises are possible.
Good point!!
I suspect that audio devices are LTI only when theiy are in their "steady state" condition. But, what happen before to reach the steady state? My test with a single pulse tries to analyze that situation. Most of the answers on what we hear could be hidden in this "starting" instant.
The use of swept sine signals doesn't allow to see what happen during this phase.
I'm aware of none that are L (linear) - perhaps the Halcro is the closest approach to linear when considered in the frequency domain.
As for time-invariant, a class A amp gets cooler under load and class AB/B gets hotter. So both would possess thermal memory.
I think you are right, luigi. It is the timing of the initial transient 'instant' that might well be compromised enough that the ear hears it somehow, as compromised or unnatural. Of course, test signals should not have 'unlimited' bandwidth. It should be constrained to about 30KHz or even less. I prefer 30 KHz, because it tracks the rise-time of the best sources that I have ever found, and I think will compare to the best SACD rise-times that are practical.
The transform of the frequency response to an impulse response works very well as long as the system is linear. Exciting an amp with an impulse will move it through all of its nonlinearities which may impact the shape in many small ways. These are usually easier to see in the frequency domain than the time domain, but you need to check at different levels. And even a chirp has a longer time to heat up junctions than an impulse.
For speakers an impulse doesn't provide enough signal for a good signal to noise on the measurement so a chirp helps a lot.
This pulse stress measurement really begs for a differential measurement between the input and the output (like a settling time measurement). With Praxis its pretty easy to do since Praxis has differential probes and a resistive divider to match the output to the input is pretty simple (less so if you are obsessed with resistor nonlinearities). The time delay through the amp will show but the increased resolution will help show differences between amps.
One aspect of this we don't discuss much is the non-ideal nature of the input circuit. Driving an amp from a 50 Ohm source will hide the cap change with level and other stuff that happens at the input of an amp. This should be included inside the measurement loop, possibly with a higher source impedance.
Obviously different loads will impact it as well. Real loudspeakers may do all kinds of interesting things to the output of the amp under these stresses, especially since they are definitely non-linear with amplitude.
With Praxis you can use a .wav file (Luigi, can you send it to me?) to drive the system or use the impulse stimulus to drive a one shot with a passive filter to create the pulse and then derive the CSD (cumulative spectral decay) from the impulse. The increased dynamics and time resolution might make more information visible. I believe your point is that the closer these difference signals are between amps the more they will have the same sonic character.
Is there a rate of repetition that will not degrade the accuracy of the test? I know its a one shot test but we aren't talking nuclear testing here. There is a cycle time that would be legitimate. Its also possible that some low frequency repetition may show a supply or bias change that would not be visible either with one shot or continuous stimulus. Perhaps something like a rhythm track?
This certainly opens the gate for more new things to look at. The plot Luigi provided suggests he is sampling at 48 KHz. I suspect a higher sample rate would show a lot, or at least make for an interesting picture.
I have been using Praxis for years to look at speakers. More recently using it to look at jitter but now amplifiers. . . It has been a great investment.
For speakers an impulse doesn't provide enough signal for a good signal to noise on the measurement so a chirp helps a lot.
This pulse stress measurement really begs for a differential measurement between the input and the output (like a settling time measurement). With Praxis its pretty easy to do since Praxis has differential probes and a resistive divider to match the output to the input is pretty simple (less so if you are obsessed with resistor nonlinearities). The time delay through the amp will show but the increased resolution will help show differences between amps.
One aspect of this we don't discuss much is the non-ideal nature of the input circuit. Driving an amp from a 50 Ohm source will hide the cap change with level and other stuff that happens at the input of an amp. This should be included inside the measurement loop, possibly with a higher source impedance.
Obviously different loads will impact it as well. Real loudspeakers may do all kinds of interesting things to the output of the amp under these stresses, especially since they are definitely non-linear with amplitude.
With Praxis you can use a .wav file (Luigi, can you send it to me?) to drive the system or use the impulse stimulus to drive a one shot with a passive filter to create the pulse and then derive the CSD (cumulative spectral decay) from the impulse. The increased dynamics and time resolution might make more information visible. I believe your point is that the closer these difference signals are between amps the more they will have the same sonic character.
Is there a rate of repetition that will not degrade the accuracy of the test? I know its a one shot test but we aren't talking nuclear testing here. There is a cycle time that would be legitimate. Its also possible that some low frequency repetition may show a supply or bias change that would not be visible either with one shot or continuous stimulus. Perhaps something like a rhythm track?
This certainly opens the gate for more new things to look at. The plot Luigi provided suggests he is sampling at 48 KHz. I suspect a higher sample rate would show a lot, or at least make for an interesting picture.
I have been using Praxis for years to look at speakers. More recently using it to look at jitter but now amplifiers. . . It has been a great investment.
John, I'm with you. The pulse I use is band-limited to nearly 24 KHz: I agree with you that 30 KHz would be better.
Martin,
I completely agree. One of my first mistake in "copying" the sound of a good sounding amp has been caused by the math: I thought "if I copy its transfer function I will get the same impulse response". Completely wrong!! It works only on ideal circuits. The interaction with power supply and layout parasitics capacitances and inductances completely destroys this math.
I completely agree, again.
Interesting topic!! I'm more confident with electrostatic loudspeakers, since I spent the last 20 years with them, but I worked with a dynamic speaker for a sub to add to my speakers. Well, according to me, direct driving of dynamic speakers offers a lot of benefits. The acoustic pressure wave is a nearly perfect copy of the cone acceleration; the cone acceleration is function of the drive current: have you ever looked at the current distortion of an amp when driving a real loudspeaker and not a resistor? You will become crazy!!
Well, I got very good results by working on the pulse shape of the driving current...
Yes, please PM me your email.
Many good questions!!
Repetition rate: I think it could depend on the time constants of the circuits; if you uses a repetition rate that allows to completely discarge all the capacitors and inductances in the circuit, this could be fine.
Rhythm track: low frequency is where the benefits of the pulse is more obvious. When you listen to an audio system, the first impact is with low frequencies; many times the defects are addicted to the room, but in many cases this is not the root cause. In my mind the problem it's a wrong energy in the drive of the loudspeakers; energy it's the area under the pulse (pulse integral). When you have booming low frequencies, try to shorten the pulse width and you will get more punch. In the low frequencies, where the energy is stronger, some defects in the power supply will affect more the sound, so your suggestion of a rhythm track sounds very interesting.
Used impulse was generated bandlimited to 24kHz (not to infinity), so here cannot be any higher harmonics as 24kHz...
In a philosophical sense, you can't even reproduce a 1kHz squarewave without infinite bandwidth. But that's irrelevant to audio. All signals that are of interest are inherently bandwidth limited. And if I measure the impulse response of an amp (a loudspeaker is different) at t1 instead of t2, the response is the same (unless I deliberately construct something nonrepresentative like making gross temperature changes between measurements).
One of the reasons I asked for definition is that "linear" has different meaning in different contexts. A square law transfer characteristic is not linear, yet can comport to LTI used in a control system sense.
One of the reasons I asked for definition is that "linear" has different meaning in different contexts. A square law transfer characteristic is not linear, yet can comport to LTI used in a control system sense.
And you think his different pulse responses won't show different harmonic spectra?
Hmmm, invalidated by others!
Last edited:
Luigi,
One thing I have been wondering about and that is what you are looking for when comparing the pulse response from one amp to that of another one? I guess you don't look at absolute level as that's a matter of absolute gain and is ultimately corrected by your volume control.
But, are you looking for rise/fall times/overshoot differences? And how much difference do you accept for concluding the amps will sound the same (or not)?
jan didden
Jan,
I make the comparison with a measurement at the same levels; in my experiences the two amps will sound exactly the same, when the two pulse are exactly the same. Even small difference in the attack/decay time are audible, especially in the low part of the spectrum.
- Status
- Not open for further replies.