PDA

View Full Version : Earthworks M30 vs ECM 8000 side-by-side test for Inguz DRC



muski
2007-09-27, 23:00
I took measurements today with both of the mics, using a Sound Devices USBPre audio interface. Very interesting results!

The first chart shows the freq response of the left channel measurements (Due to differing sensitivities the sweeps were at slightly different levels. So I shifted the ECM8000 up about 4db for the sake of comparison).

The second shows the phase response (built with the drc Octave scripts).

The third shows the Test_Convolution of the normal filters computed with a flat target curve and each mic's calibration file (unshifted traces). The ECM8000 filter looks a lot like the mic cal file profile. (I am still suspicious/baffled about drc's use of mic cal files).

The fourth chart shows the frequency response of the two mics -- ie a plot the calibration files. I had the ECM8000 calibrated by a third party and Earthworks provides a calibration file for all their mics for an extra $50.

It's interesting to see that the ECM8000 does mostly OK in the lower freq, but not so well above 1KHz.

Only had time to get these measurements done and have yet to do a listening test. One day I'll try same test with an Radio Shack SPL and also compare the Tascam US-122L to the Sound Devices USB Pre.

muski

Chinanico
2007-09-27, 23:17
It's interesting to see that the ECM8000 does mostly OK in the lower freq, but not so well above 1KHz.

muski

Very interesting! Thanks for sharing. Not a good news for ECM8000 users like me (and like most here I guess). I assume that the response in the treble would be more sensitive to the exact position of the mike than in the bass. Do you think this could account for some of the differences or did you manage to position them the very same?

Anayway there is still something positive about it: this would mean that if I take the path of Audiolense to take measures and generate filters for below 200Hz correction only, I shall be fine. the more I listen to filters and the more I thnik I cannot live without the low frequency corrections, but that in the upper frequencies this is more a matter of the sound being different, sometimes better, sometimes not, depending on the music (definitly clearer and with more lisibility, but sometimes too much on the dry side...)

nuhi
2007-09-28, 04:16
muski, so it is confirmed, without a calibration the ECM8k is almost useless :(

Tell me one thing please, in the first pic, which calibration file for the ECM did you use?
I hope it was none (zeros) so that I can adjust mine.

Very helpful indeed, so we do have hi freq issue. I say we because I have the same recorded response with my ECM.
Also I own Bryston 4BSST...could it be the cause hmm ;)

)p(
2007-09-28, 04:41
The ECM8000 filter looks a lot like the mic cal file profile. (I am still suspicious/baffled about drc's use of mic cal files).

Almost exactly...like it was the target instead of flat...


I see something similar with my mic-cal which has a gently rising slope from 5k upwards to 20k of about 5 db. I also see that reflected in the filter but too a lesser extent then in yours.

peter

tonyptony
2007-09-28, 06:14
muski, so it is confirmed, without a calibration the ECM8k is almost useless :(

This gets more and more interesting. I sent my ECM8000 out as well to get calibrated. I'll post my cal curve as soon as I get it back.

muski
2007-09-28, 07:10
I assume that the response in the treble would be more sensitive to the exact position of the mike than in the bass. Do you think this could account for some of the differences or did you manage to position them the very same?

I used a mic stand and was very careful to get the tips of the mics in the same place. From the prep log file, it looks like maybe it was 1cm different in the R channel (though these plots show the left channel, which are the same).

The M30 prep.log:

Processing left measurement (m30left.wav)...
Right channel seems to be the sweep
Impulse peak at sample 444 (3.43m, 11.24ft)
Deconvolution: left impulse done.

Processing right measurement (m30right.wav)...
Right channel seems to be the sweep
Impulse peak at sample 444 (3.43m, 11.24ft)
Deconvolution: right impulse done.

The ECM8000 prep.log

Processing left measurement (ecmleft.wav)...
Right channel seems to be the sweep
Impulse peak at sample 444 (3.43m, 11.24ft)
Deconvolution: left impulse done.

Processing right measurement (ecmright.wav)...
Right channel seems to be the sweep
Impulse peak at sample 443 (3.42m, 11.22ft)
Deconvolution: right impulse done.

muski

muski
2007-09-28, 07:15
Tell me one thing please, in the first pic, which calibration file for the ECM did you use?

The first graph is just the Impulse_Response_Measured, so there is no mic calibration applied (ie zeros). In fact, I can't figure out how to generate mic-cal adjusted graphs of the sweeps...

muski
2007-09-28, 08:25
Here are the same first three graphs for the right channel -- just to double check things.

Freq response is very similar, though the phase plot is slightly different. Maybe the ECM8000 is actually ok in terms of capturing phase information.

nuhi
2007-09-28, 08:41
muski, oh ok.

While adjusting the curve to your new comparison I was wondering are those dBs on the phase response actually degrees or it's more like divided somehow?

I'm asking because in the default ecm8000 there is a huge difference, like 3 times stronger correction in phase (they use 20, your shows 7 for example).

Maybe this graph of yours is 3 to 1 or something when it comes to phase?
I would want it to be 1:1, just asking if you know.


And let us know how does it sing, you haven't commented yet, don't tell me you're not satisfied?
Maybe at first you'll miss the 3khz bump and it may sound more polite and boring but when you get used to the real flatness it's great.


Btw for those experimenting try this ecm8000.txt correction:
0.0 0.0 10.0
20.0 0.0 9.0
23.0 0.0 0.0
158.0 0.0 0.0
200.0 0.0 0.0
2000.0 0.0 0.0
3000.0 0.0 0.0
10000.0 7.0 -7.5
15000.0 3.0 -4.0
22050.0 0.0 -3.0

muski
2007-09-28, 09:18
muski, oh ok.

While adjusting the curve to your new comparison I was wondering are those dBs on the phase response actually degrees or it's more like divided somehow

No, it's degrees (-180->0->180). (BTW, I have no phase information in either of my mic cal files). It seems like these phase response plots vary a lot depending on the amount of smoothing done. I'll have a look at the other Octave phase plots.

From the DRC documentation, it's not clear how much benefit there is to modifying the phase information. Also, it looks like you need to modify the .drc files to make sure all the MC flags are set correctly.

From http://drc-fir.sourceforge.net/doc/drc.html:


Starting from version 2.0.0 DRC lets you specify the phase for the target post filter stage. Phase specification should be placed after the amplitude specification and should be expressed in degrees. Following the example above:
0 -40 0
18 -20 45
20 0 90
20000 0 180
21000 -40 90
22050 -100 0
If not specified a value of 0 is assumed. Setting a phase different than 0, i.e. flat, is useless within normal HiFi systems in almost all circumstances. Furthermore the phase specification is used only if the PSFilterType is L, else any phase specification is wiped out by the minimum phase filter extraction.


Looking forward to doing some listening tests tonight!

muske

muski
2007-09-28, 14:39
I used the "Sweep (with EQ in L channel)" test tone and, using the M30 microphone recorded sweeps for both the M30_normal_flat and ECM8000_normal_flat filters. Below are the plots of "Impulse_Response_Measured" and phase response (1/6 octave smoothing).

Since the M30 is pretty close to flat even (see the first post), the fact that these are not mic cal adjusted shouldn't make much difference.

It is interesting that the ECM8000 plot, though not without some issues, does looks flat-ish (ie at least it doesn't have the huge hump like the ECM8000's mic freq reponse plot). This suggests that the mic cal files are correctly processed in drc?

I am surprised by the phase behavior of both filters in the low freqs.

muski

opaqueice
2007-09-28, 17:12
I am surprised by the phase behavior of both filters in the low freqs.


That phase response may not be as bad as it looks. The point is that if a filter is linear phase it's perfect (just as good as 0 phase shift), because linear phase is simply a time delay. That's because if f(t) = sin(\omega t + \phi), when \phi -> \phi + c \omega for some c that doesn't depend on \omega, it's the same thing as shifting t -> t + c.

On a plot like that linear phase means a line of a constant slope, but of course the lines will wrap around from -180 to 180. That's more or less what it looks like at least up to 2kHz or so. And I assume this plot has been smoothed, which might account for the apparent change at high frequency.

nuhi
2007-09-28, 17:24
It is interesting that the ECM8000 plot, though not without some issues, does looks flat-ish (ie at least it doesn't have the huge hump like the ECM8000's mic freq reponse plot).

There are 6dB jumps all over the high frequencies, that's too much (3dB is 2 times louder).
It looks even worse because of all the problems in the lower frequencies. But that is due to the correction filter options probably. Looks worse but I bet it sounds better.



This suggests that the mic cal files are correctly processed in drc?
Sorry what do you mean by that, didn't you say that the mic cal files weren't applied?

muski
2007-09-28, 18:05
Sorry what do you mean by that, didn't you say that the mic cal files weren't applied?

The mic cal files were indeed applied to create the two filters. The M30 mic cal file was not applied to the measured results shown above, but as the M30 is reasonably flat it shouldn't make much difference.

So, in summary, the mic cal files were applied to create the filters, but not to measure their effect.

nuhi
2007-09-29, 03:22
Then the ECM8k correction file isn't good.
For example it has 10.5dB attenuation in the 10khz area while your new graphs show that 6 is enough.
Maybe it would be better if you use strong correction, maybe there was some approximation by the normalization.

If you find the time please try this (attached), I made it by looking at the difference in your graphs between the m30 and ecm.

Chinanico
2007-09-29, 20:19
If you find the time please try this (attached), I made it by looking at the difference in your graphs between the m30 and ecm.

I might try that. Actually I was thinking, as more and more people get their ECM8000 calibrated, but others don't... (or can't, for me in Shanghai I wouldn't know where to go), would it be possible to build so kind of "database" of the calibration files and compare them. Maybe we will see that there are some patterns that could be used to extrapolate a theoretical "average calibration file" that we could use by default if we didn't have had calibration (and that might better the default from DRC)... or maybe we will see that there is no way but to have a calibration.

I know some of you guys already published files, but we could start to collect them more formally, if you agree.

muski
2007-09-29, 23:31
Recorded sweeps using both of the Tascam US-122L and the Sound Devices USBPre audio interfaces today. The Earthworks M30 mic remained in exactly the same spot. The Octave plots MRFDWSmoothed.jpg and PRFDWSmoothed-1-6.jpg are shown below.

muski
2007-09-30, 07:49
I might try that. Actually I was thinking, as more and more people get their ECM8000 calibrated, but others don't... (or can't, for me in Shanghai I wouldn't know where to go), would it be possible to build so kind of "database" of the calibration files and compare them. Maybe we will see that there are some patterns that could be used to extrapolate a theoretical "average calibration file" that we could use by default if we didn't have had calibration (and that might better the default from DRC)... or maybe we will see that there is no way but to have a calibration.

I know some of you guys already published files, but we could start to collect them more formally, if you agree.

Here is mine...

opaqueice
2007-09-30, 09:25
Here is muski's calibrations file (vertical axis in dB, horizontal is log base 10 of frequency). The second plot is smoothed by averaging over each consecutive set of 10 values.

It looks kinda like the plot Behringer gives here, in the specs, only.... not as flat and with a much bigger hump at high frequency :-).

http://www.behringerdownload.de/ECM8000/ECM8000_C_Specs.pdf