Speaker measurement: Live capture and calibration using amplifier speaker output. Is it possible with REW?

[p-we]

Hello REW community,

I have been a faithful customer of Biil Wazlo's Liberty instruments products through the years but Praxis is 32 bit, getting fussy with new hardware and constant changes in the Windows environments. So trying REW and it looks cool. I have also searched this topic and couldn't find a hit.

Proper calibration is of utmost importance to differentiate what the speaker is doing from the signal chain. I have made a CAL file from a measurement of the output of my measurement amplifier, and it works, but IMO one measurement reference is not representative of all scenarios. Any change in load impedance or setting of a volume button can alter the amplifier output. I will be paranoid if I don't know for sure if any artefact in the analog chain is not accounted for. So I believe the signal being fed into the speaker must be probed live at every measurement and calibrated into the measurement data. I have tried to mimic this by using the feedback loop with a 50k series resistor connected to the positive speaker terminal, use the same output source for the signal and the loopback, using the "Use loopback as cal and time reference" setting, and using another input as the loopback. Alas the loopback signal is not picked up right and not useful.

Does anyone know how to use the loopback functionality to do this? It would really be great to have 2 loopback paths, one for the soundcard (internal) and one for an external analog signal calibration where the analog loopback is assumed to be the same as source output. I suppose I could create a new cal every single time I make a new measurement but that would be a pain.

Thanks,
Wes

 

[John Mulcahy]

Seems redundant. If your measurement amplifier has decently low output impedance you are only compensating for the amplifier bandwidth limits, which you could calibrate out by measuring the signal chain once. If it has high output impedance (not a good idea for a measurement amp) you'd be getting back EMF from the speaker at the loopback point which wouldn't represent what the amplifier was trying to reproduce.

If you were determined to use the amp output as a loopback signal you'd need a proper voltage divider, not just a series R, with the shunt across the interface input significantly lower than its nominal input impedance and with an appropriate overall divide ratio, plus zener protection for the input to make sure it can't get blown up accidentally.

 

[p-we]

John, I find it to be special when the developer is the first to reply. REW is amazing and quite the achievement. My gratitude to you for making it easily accessible. I'm trying to see if I can migrate to REW.

Probing the speaker output is quite common, probably ubiquitous, in dedicated speaker testing systems. I know the legacy of REW is of course room acoustics. I don't not know about the commonality of a voltage divider probe as this could be in the hardware of some of the legacy systems out there. Liberty Instruments have traditionally used balanced input probes with two series resistors to each terminal. So I tried this realizing that my USB soundcard has TRS balanced inputs and outputs. This works much better than a single resistor unbalanced setup I tried before, although 50k is too much attenuation for typical low SPL measuring levels. So it is now successfully making a CAL at every measurement by using the same output for both signal and loopback and feeding the balanced input probes into the loopback input. I noticed I can't use a pre-existing CAL for the sound card as it is a duplication of a part of the loopback signal. What is unfortunate that I can't do both a +/- "Use loopback as cal.." and "Use acoustic timing reference" at the same time.

Regarding amplifiers, you are right that using very high quality reference output amplifier could make live probing unnecessary. While especially in the old days, even today a quality amp with a stated damping factor of 100 at 1kHz could have significant skewing of the linearity depending on the impedance load, especially when the impedance load goes into negative phase. Then there is the practical aspect of using a valuable reference hi-fi amplifier in a development or portable environment. Nowadays a small class D amps are quite good, but far from perfectly stable under variable loads and probing the speaker output guarantees the amplifier's fingerprint is out of the equation. I could be wrong but I think it is more critical when designing a speaker, compared to room acoustics, to not underestimate the dependence of driver behavior on amplifier behavior under variable load conditions.

Thanks again

 

[p-we]

John, I've been thinking about your response and how back EMF figures into it. I think back EMF phenomena, which is Z load dependent, is always present to some degree and the main reason why amplifier outputs as a general rule do not have consistent linearity in the audible FR range. I'm curious, so I'm going to make some measurements of the amplifier's I have access to and compare resistive loads to a typical bass reflex impedance load. See what comes of it.

 

[p-we]

I've compared the output of two Class D amps: Hypex UcD180HG vs.TDA7489E. Each graph has 2 plots with 8.2 ohm resistive load and a 4 ohm bass reflex speaker. The measurements a were done with a +/- 5 watt stimulus. There is a difference but indeed, not huge. What I don't understand is what is causing the LP filter effect at 9kHz -3 dB. The soundcard cal is within 0.2 dB to 40kHz and does not influence these measurements at all.

 

[p-we]

The measurements were done using the 50kohm balanced probe to the speaker outputs as discussed above with a soundcard cal file and without a loopback. Also 8.2 ohm is a mistake, It was a 3.3 ohm resistor, so overall closer in nominal impedance to the real speaker

 

[John Mulcahy]

The effect of your 50k series Rs will depend on the input impedance of the input stage. The capacitance of the input may contribute some roll-off. Try lower series R to see what effect that has.

 

[ddude003]

It might be interesting to also add the impedance overlayed onto the charts...

 

[p-we]

Investigating the HF attenuation issue, I opened up the Liberty Instruments 'Audpod' hardware interface to find both external probe inputs are indeed shunted by a 4.7k resistor immediately at the junction as you said it would have to be. Since they recommend 47 to 50k series resistors, a +/- 10:1 ratio is the target it seems. So I placed 5k shunt resistors creating a voltage divider for each balanced input at distal end of the series resistors and not at the other end of the cable. While this does improve things, the HF attenuation is still there, possibly to a lesser degree. I don't think there would be such a huge difference compared to the scenario of having the shunt resistors at the other end of the balanced cable or not. Will try smaller series resistors

I'd love to crack this to be able to accurately measure/calibrate amplifier output. Then possibly measure impedance through amplified probes as well

 

[p-we]

Did some searching online. Found systems using much lower shunt resistance +/- 1kohm and the 1:10 ratio. Have now tried with 9kOhm series with 1k shunt. Distortion measurements now look to be based on true amp spec's, previously quite high. This is the TDA7498 connected to a real speaker

 

[p-we]

This was supposed to be about the comparisons of amplifier FR under different Z loads, but now also includes understanding how to get the probes done right in REW. Regarding the variability of amplifier FR, I think there is merit in doing real time calibrations even if the effect size is small.

Then using REW with the balanced probe setup below is causing a resonance at 30kHz when loaded with an inductance of a real speaker. The resonance does not appear with the resistive load but the 2 curves are still different. This resonance is an artefact and does not happen when measuring with Praxis. REW and Praxis match well in the lower frequency ranges. I wonder if the soundcard (Behringer UMC404HD) is causing. In Praxis the probes are connected to a proprietary interface and not directly not the card.

 

[ddude003]

Have you seen this article? https://www.audiosciencereview.com/...r-amplifier-module-analysis-and-review.27852/ Does it help explain any of what your seeing with your testing... I am no amp expert and it seem that many class D amps perform poorly at high frequencies... Maybe why they are so popular in powered subs...

Wondering how this amp would respond to a speaker with a reversed impedance map, diving below 2 ohms at 20khz, like an Martin Logan ESL...

 

[p-we]

ddude003, thanks for responding.

When looking at various Class D reviews, including Hypex, while not perfectly linear above 20kHz, do not replicate this kind of huge resonance with a big peak. Looking at the curves I made with Liberty Instruments Praxis using their proprietary audio interface hardware (see below), there is no resonance peak even when connected to a real speaker with an inductive load at high frequencies.

I believe this is a measurement artefact and not actually from the amplifier being measured. I am curious what another soundcard would look like using the same probe circuitry with REW.