Need advice - JBL 305P - DSP room correction - REW and DRC-FIR - results not that great

[asteroth]

I've measured again, microphone in listening position, 2.x meters away from the speakers, at listening volume levels. Preamp on the interface was at max volume.
I see that the level is low, but what can I do?

 

[Keith_W]

This is why I like looking at REW measurements, every now and then you come across something really unusual. Never in my life have I seen a measurement that looks like this. Look at that, four peaks on the ETC that are equal in SPL and the exact same shape and equal timing between the peaks!! And the same thing on the impulse response! The measurement of the right speaker looks the same.

And look at all that comb filtering in both measurements!!!

All I can tell you is that this measurement is uninterpretable and unsuitable for DSP correction.

As for what went wrong ... this is a real head scratcher. It looks as if your speaker is producing four impulses instead of one. Or the maybe the first peak is the impulse and it is followed by 3 reflections which are perfectly spaced ... which seems extremely implausible. And I don't think that microphone clipping would produce an ETC like this.

Given your microphone problems I wondered if it could be poor SNR. So I used my usual trick to examine the noise floor - extend the waterfall to 1000ms and the vertical range downwards, then "normalize to peak of each frequency". And this noise floor looks weird as heck. What is that peak at 240Hz?? Sometimes, things like spinning computer fans can produce a peak like this, but 240Hz = 14400rpm. I can't think of anything that spins at that speed.

I did notice that you are using a 44.1kHz sampling rate and you are using a Behringer UMC204HD. That product page says that its sample rate is 192kHz. Can you please check that you set the sample rate on your interface at 44.1kHz? And have you checked that the sample rate in REW matches the sample rate on your interface? Even then for the life of me I can't imagine why a mismatched sample rate should produce a measurement that looks like this!

Otherwise, I am out of ideas. I am going to have to call in some help for this one. @OCA? @3ll3d00d? @JohnPM?

 

[Keith_W]

Oh wait, I figured it out!

What we are looking at is the sum of four sweeps with no timing reference. No timing reference means that REW can't sum the four sweeps properly. That is what is producing the train of four impulses. And because the impulses are delayed relative to each other, that is what is producing the comb filtering! In other words, you are looking at a measurement artefact caused by improper summation of four sweeps.

So you need to:

1. Use a timing impulse with all your measurements. If you are having trouble with "low level", try using a loopback timing reference. You will need to connect an unused output on your interface to an unused mic input with an appropriate cable. Make sure you turn off 48V Phantom power on that input. Maybe your interface supports software loopback, I don't know.
2. Load a microphone cal file.
3. Use REW's soundcard linearisation procedure.

The only question that remains is why the microphone sensitivity is so low. I can tell you from my own experience that the ECM8000 has fairly low sensitivity compared to my Earthworks M30.

This is a comparison of my Earthworks M30 (green) vs. ECM8000 (purple). Both have appropriate cal files loaded. Both are nearfield measurements of the tweeter. Both have the exact same settings on the interface. Notice that the ECM8000 has about 30dB less volume than the Earthworks, and the shape of the curve is different.

The published sensitivity of the M30 mic is 34mV/Pa, and the ECM8000 is about 8-10mV/Pa. The 20-25mV/Pa sensitivity translates to 20 * log10 (20/1000) = 34dB, so my measurements are in line with what you would expect. Also the ECM8000 is a rather cheap microphone, I would imagine that QC would be quite poor, so some batches may be less sensitive than others.

If your interface does not have enough gain, I can imagine that you would struggle with microphone sensitivity under certain situations causing REW to complain of low measurement volume. I don't normally recommend that you measure too loud since you will likely get a different freq response. But in this case, I think that you have no choice.

I am usually reluctant to recommend equipment purchases, but I think you may have to borrow another microphone and compare the measurements. Just for a sanity check.

 

[JohnPM]

What we are looking at is the sum of four sweeps with no timing reference. No timing reference means that REW can't sum the four sweeps properly. That is what is producing the train of four impulses.

Close. You don't need a timing reference to make multiple sweeps, but the input and output devices need to use the same clock otherwise sample synchronisation cannot be maintained. The input and output device are different here, so multiple sweeps should not be used.

 

[Sokel]

@asteroth, I had a similar problem a few years ago when my microphone lost sensitivity and I could not obtain any decent measurements. One run would be OK, and the next I would lose a lot of microphone sensitivity. I had to turn the gain on the preamp all the way up.

I went bonkers trying to diagnose the problem. The problem had to be the microphone, the cable, or the interface. I tested the cable with a digital multimeter, and it was fine. I have three interfaces, and swapping interfaces did not improve the problem. So I bought another microphone, which is why I have two ECM8000's. Then the new mic had the same problem. So now I KNEW FOR SURE that it was the cable.

I was too cheap to buy an XLR cable of the proper length, so I made up a longer cable by plugging shorter cables together. It turned out that the connector of one of the cables was not manufactured properly. When I opened it up, I saw a strand of loose wire. Twist the connector a certain way, and it would short with another wire and send signal to the ground. As you can imagine, it was a real bitch to diagnose because of its intermittent nature. All that required was to snip off the loose strand and the problem went away. It would have been a $0 fix if I didn't have to spend money buying a microphone I didn't need.

That's your DMM's revenge for not using it before using (any) cable

 

[asteroth]

Guys, I really want to thank you for your effort into writing these very helpful posts.
Now I understand better the problems that I have caused due to my lack of knowledge in this area.
I will try to buy or borrow a better microphone(although I cannot afford that M30, which looks like a gem at least for the sensitivity that it has, very useful to measure at your listening levels) and cables for loopbacks. I'll use the Behringer interface for both input and output from now on, but I need to buy longer cables first.

 

[Keith_W]

Close. You don't need a timing reference to make multiple sweeps, but the input and output devices need to use the same clock otherwise sample synchronisation cannot be maintained. The input and output device are different here, so multiple sweeps should not be used.

Thank you. I seem to recall from your experiments with USB microphones that sample synchronisation error was in the order of 100-200 us (microseconds). These samples are spaced much further than that, about 2ms apart. Would a timing reference allow the four impulses to be summed if the inputs and outputs were on different devices?

I don't know whether to be embarrassed that it took me so long to understand what happened, or proud that I managed to figure it out. In hindsight, looking at that train of four impulses ... it looks so obvious. But I was utterly flabbergasted and confused when I first saw it.

 

[asteroth]

I did a new set of measurements, 1 sweep for each channel, I increased the volume also. For some reason, this time it managed to perform the sweep on the R channel without any pops.

 

[JohnPM]

Would a timing reference allow the four impulses to be summed if the inputs and outputs were on different devices?

The input from multiple sweeps is averaged before any processing happens, based on the number of samples in the stimulus, so the input stream has to be sample-coherent with the stimulus. Multiple time-synchronised single-sweep measurements could be averaged.

 

[Keith_W]

@asteroth please pay attention to the previous replies before you do your next measurement.

1. The sample rate is 44.1kHz. Did you set the same sample rate across your entire measurement chain?
2. There is no timing reference. As per the discussion with John, if you want to improve your SNR, you can take several sweeps with a timing reference and average them. Make sure you examine each one and discard the rubbish ones before you average. Symptoms of a "bad" measurement are: inconsistent timing measurements, funny looking impulse response, and whether you hear external noise during the sweep (e.g. dogs barking, plane flying overhead, etc).
3. There is no cal file for the microphone or the interface.

4. There is a lot of pre-ringing in the impulse response. Are you measuring with DSP filters in place? Can you try removing them and see what the result looks like? There is also a very early and loud reflection at 1.5ms. Is the left speaker placed next to a wall?

Otherwise, this measurement looks much more like what we would expect to see. We can work with this.

 

[asteroth]

@asteroth please pay attention to the previous replies before you do your next measurement.

1. The sample rate is 44.1kHz. Did you set the same sample rate across your entire measurement chain? Yes
2. There is no timing reference. As per the discussion with John, if you want to improve your SNR, you can take several sweeps with a timing reference and average them. Make sure you examine each one and discard the rubbish ones before you average. Symptoms of a "bad" measurement are: inconsistent timing measurements, funny looking impulse response, and whether you hear external noise during the sweep (e.g. dogs barking, plane flying overhead, etc). There could have been some external noises, right now I don't have the loopback cable, so I decided to do a single sweep.
3. There is no cal file for the microphone or the interface. I need to buy a calibrated one and calibrate my cheap one.
4. There is a lot of pre-ringing in the impulse response. Are you measuring with DSP filters in place? Can you try removing them and see what the result looks like? There is also a very early and loud reflection at 1.5ms. Is the left speaker placed next to a wall? "DSP" in EQ-APO was off. The room is asymmetric. The left speaker is has the backwall 30cm away from the bass reflex port and the sidewall 45cm away. The right speaker has the backwall 30cm away and the sidewall is a few meters away.