accuracy of REW for full range measurements?

[terry j]

hi everybody.

After mucking about with REW with the DEQ 2496, I've moved on to play in the big league, the DEQX unit.

However, I'm getting poor results when using REW for full range measurements, and I'm very sad about it... ( ha ha )

To refresh memorys and by way of fuller explanation, I'm using an ECM 8000 microphone with the generic correction files loaded, into a Xenyx 802 ( replacement unit for the UB 802) then of course into the soundcard.

I seemed to be getting 'acceptable' graphs when measuring the results of using the DEQ 2496, but the graphs now don't look that good, as you will shortly see.


here is the graph of the verification measurement done by the DEQX unit after calibrating the speaker.





Without moving the speaker or the microphone, here are the resulkts of using REW to measure the speaker, firstly with a 20 to 20 sweep, then by 1/12th octave auto measurement.





As you can see, the plots bear no resemblance to each other, let alone the results of the DEQX verification.The rising graph is the sweep, the interesting descending one is the 1/12th octave.

I don't even know where to begin to explain it, and am really hoping that there is a simple explanation. What I really don't get is why my graphs using the DEQ 2496 at least looked in the ballpark, these look waaaaay off.

Just noticed the scales are wrong in the picture, sorry but I don't think that will prevent understanding of the problemo

lots of love
terry

[Shez]

Your purple trace looks like something is causing comb filtering. This happens when you mix a signal with a very slightly delayed version of itself. Check your routing to make sure you're only capturing the signal from the mic and not accidentally mixing it with the actual test signal.

(You can work out how long the delay is given the frequency of the first null but my brain doesn't feel like trying that right now...)

[JohnM]

You have inverse C correction turned on in REW (hence the C symbol by the trace values), it is the inverse C correction which gives that rising curve at high frequencies - only intended for use with SPL meters, so for your ECM8000 you should turn it off (in the Meter menu).

[terry j]

thank you john, must say I've never noticed the c to -1 before!! Brilliant. Sometimes had 'sensible' measurements, others not.

Thanks shez, not sure you're analysis of the purple trace is correct. Is an interesting one tho isn't it! That trace was 1/12th octave each measurement, took around 18 mins to do! Left it and went and worked elsewhere. It was done outdoors, nearest object is about 4m away ( except for the ground ), even then was on a stool so almost 2m off the ground. But I suppose you could be right, those conditions certainly qualify as 'slightly delayed'.

But Johns 'troubleshooting' is certainly the way forward for tomorrow, will let y'all know how it goes.

lots of love

terry.

[brucek]

Just noticed the scales are wrong in the picture, sorry but I don't think that will prevent understanding of the problem

I would also match up my scaling. On the DEQX they are using a vertical height of 100db, where on REW you are using about 50dB....that's a huge difference to interpret.

Also they have 50% smoothing turned on in the DEQX that you should match on the REW graph.

Don't know if the C-weight switch will help. If you used the ECM8000 file from Sonnie, then it covers from 10Hz to 46KHz, so it should over-ride the C-weight switch. If you didn't use Sonnies file, you should.....

Use auto sweep...

brucek

[terry j]

thanks Bruce

wasn't too worried about both graphs having the same scale, was more worried about the totally different nature of the responses. Today tho, just for interest and completeness have got both scales with a 60 db range. Also, I am using the ECM cal file.

Unfortunately John, the results haven't changed.




This graph shows the response both with and w/out c weighting. Bugger.

The DEQX graph I redid today looks the same as yesterday, essentially flat thru the graph.

Also, forget the DEQX graph today, although the site accepted it y/day, it won't today. It is too big, and stupid me doesn't know how to re-size it. Can someone give a real quick explanation? Currently I just use print screen and paste into paint. Once there, if I change number of pixels it truncates the picture rather than resize. To overcome that problem in the above graph, I used the little symbol to the left which then allows me to size it, but of course miss out on all the data on the rest of the page.

Strangely enough, I think I can hear the frequency sweep change during the REW measurement. What I mean is, I feel I can hear the rising frequencies actually get louder and the resulting graph is in fact an accurate representation of what it hears, in other words the fault may not lie in the measuring side of things - which was my first thought - but may lie in the output side. Is this possible??

Maybe I need to re load the program?

Is there, to anyones knowledge, any reason why REW can't be used for full range measurement? I sure hope it can cause I reckon it's fabulous.

Interesting Bruce that in future versions of REW the only option will be sweep. Has it indeed been found that the other options are inaccurate? not needed? something else? In the graphs a couple of posts above, the 1/12 step very closely followed the sweep up to whatever frequency it was, but diverged after that. Funnily enough, above that point BOTH measurements produced 'garbage', just diametrically opposed!!

lots of love

terry

[brucek]

Maybe I need to re load the program?

Not likely, it's just an exe.
Is your soundcard.cal file good from 10Hz to 20000Hz?
If you connect a cable from the LINE_OUT to the LINE_IN right channels and do a full sweep measurement with the soundcard.cal loaded and C-weight unchecked and microphone cal removed, is it indeed a perfect flat line response? It should be.

any reason why REW can't be used for full range measurement?

Certainly it can, as long as you have a proper full range soundcard.cal and a microphone.cal loaded.
See my thread here. Look at the graphs.. your loopback measurement should look like the red line..

Has it indeed been found that the other options are inaccurate? not needed?

Not needed. If you need individual frequencies, you can use the internal signal generator.

brucek

[terry j]

thanks bruce

you may know by now that I managed to get the soundcard calibrated successfully. I remember having trouble originally, but saw a flat line once in my testing back then and moved on.

All the speakers have come back inside, the mic has just been thrown in front of the speaker any old how, so the actual graphs won't match, except 'in kind'.

So, to set the scene with new graphs, because of the above changes, I was getting 'crazy' graphs remember



On your advice and suggestion, went into 'retrace steps from the beginning' mode
and reset the soundcard. Then I decided, as it was all set up, to measure the Xenyx 802 as you had when you did the UB 802.

And guess what I found??? The eq knobs in the mid and treble had been turned way up!!

I hadn't touched them, so I never even thought to check ha ha. Maybe one of my daughters, or even maybe it got kicked, but it sure would explain that huge rising response from about 1k wouldn't it ha ha.



there we go, mystery solved and lessons learnt.

lots of love

and really do appreciate your patience and help,

terry

[brucek]

And guess what I found??? The eq knobs in the mid and treble had been turned way up!!

Yeah, good catch. It was those pesky knobs that precipitated my response check. There are enough knobs on that mixer to operate the space shuttle and they can be easily ignored. It's important to check them once in a while I guess. It's so easy to get one out of its detent and it will completely throw off a measurement. , you even found one detent wasn't calibrated.

The graph looks good now...

brucek