Help wanted in interpreting REW

[Jochie]

Hi, beginner here. Would love some help interpreting REW data.
The listening part of my room is: 380cm (150inch) (front to back) cm x 350cm (138inch) (side to open floor plan, towards my back when pic was taken). Gear: pc>minidsp flex> svs pb2000 XO 80Hz BW24/dB and Apollon ET7040>Dali Opticon 6 XO 80Hz BW24/dB. After some unsuccessful tries aligning the mains with the sub via the Alignment Tool, I did it this way by ear:

https://www.reddit.com/r/audiophile/comments/1eh0izn/21_time_alignment

All the measurements are taken from MLP (I push the couch against the back wall and set the tripod on the ground with the Umik 1 pointing at the stereo, also using the 0 degrees calibration file). Dirac, when on, has the curtain set at 250Hz.

Questions:
- Why is the 'sub only' measurement so low in SPL? I only change the frequency range to 20-200, and mute the left speaker (right speaker gives the dither)
- Does the 10.5 ms delay look alright compared to no delay? To my ears, the bass sounds more tight and coherent. I don't understand the phase readings yet, and I feel ChatGPT is bsing me.
I hope I made the correct measurements, but here to learn, so feedback is much appreciated. Thanks!

 

[Keith_W]

I took a look at your measurements. The TLDR version is: there is a major problem with your measurement and REW gives you some hints that something is way off. The tricky part is to diagnose where you went wrong and what you need to do about it.

The first hint: the sub timing is WAY OFF. This is 100% a measurement artefact because DSP can not add so much delay, and your house is not so big that your sub can be located 620m away.

The second hint: the sub measurement also looks bizarre. The legend says that the XO is 80Hz, but this sub is clearly not low passed at 80Hz. It looks more like 20Hz or even lower, with a very steep slope.

The third hint: very weird looking impulse response measurements. Look at that step response. At that time scale, (-100ms to +500ms), the step response should look like a flat line with a small perturbation caused by your measurement.

The reason why can be seen in RT60 Decay: poor signal to noise ratio. This is a measurement of your L speaker with sub. If you look at the waterfall, which I have extended to 500ms, you will see that the signal is the mountain, and the noise floor is the "flat" floor where the mountain rises from. In all listening rooms, the noise floor rises at low frequencies. This is because rooms act as a low pass filter, and many noise sources are low frequency. To make matters worse, this is exactly in the same frequency band where output from speakers starts to drop.

Please examine this measurement yourself. Select this graph, then go to "RT60 Decay". Click generate, then move your pointer at various points in the waterfall. The upper graph (SPL vs. time at that specific frequency) updates as you move your pointer. Observe that at high frequencies, the SPL drops quickly and meets the noise floor after a certain period of time. Now look where I have placed the pointer - there is barely any signal above the noise.

I have to say that the shape of the noise floor itself is pretty bizarre. I have never seen a "convex" looking bass mountain like that, it normally looks concave. This artefact is present in all your measurements, so it's not something intermittent (plane taking off, neighbours using power equipment, wife using hair dryer, etc).

The REW eBook gives you some pointers on how to improve your SNR. See Book 2, Part 2, page 18 onwards. Do everything you can, shut the windows, turn off air conditioning, etc. Take a few 512m sweeps. Examine each measurement and reject the rubbish ones.

 

[Jochie]

Thank you Keith. I tried to minimise noise floor by measuring at night and turning off as much as possible, but you gave me enough pointers to go and figure out what went wrong. I'll report back in a couple of days.

 

[RandomEar]

Thank you Keith. I tried to minimise noise floor by measuring at night and turning off as much as possible, but you gave me enough pointers to go and figure out what went wrong. I'll report back in a couple of days.

You should also increase the SPL you measure at. 85 dB is a common value. This directly increases your SNR by 10-15 dB compared to the current measurement.

 

[Keith_W]

I don't like increasing the SPL of the sweep. Loudspeakers are nonlinear devices, and if you play them too loud you will get a different frequency response. There are lots of reasons why - distortion, voice coils heating up, drivers hitting excursion limits, etc. Measurements should always be taken at normal listening volume. SNR can be improved by taking a longer sweep, or by taking multiple sweeps and averaging them.

 

[janbth]

All the measurements are taken from MLP (I push the couch against the back wall and set the tripod on the ground with the Umik 1 pointing at the stereo, also using the 0 degrees calibration file).

What kind of tripod is this? Is the UMIK-1 close to the floor? Ideally you should position the microphone where your head will be (if you are doing a one-point measurement).

 

[Jochie]

What kind of tripod is this? Is the UMIK-1 close to the floor? Ideally you should position the microphone where your head will be (if you are doing a one-point measurement).

I'ts a cheap microphone-stand that can be bought with the Minidsp, it can be extended up and down. I used it to measure at ear height at MLP.

 

[janbth]

You could also try to use MMM (Moving Microphone Method) as an alternative to sweeps, just to check that your measurements are OK.

Link: https://www.audiosciencereview.com/...phone-method-mmm-for-dummies-using-rew.51333/

 

[Jochie]

@Keith_W would you be so kind to take another look?
I've been sweeping 1m (3 sweeps per object and deleted the "worst") and haven't gotten the weird artefacts and errors from REW, thanks for the tip. I used the Alignment Tool, which gave me 8.2ms delay, added that and but I'm having trouble reading the graph - to my eyes the delay worsens the response at around 45Hz. Also, when I used the alignment tool again, it gave a different result: (in REWs defence, probably because I made an error somewhere)

Question: if I understand this right, REW advices me to delay the mains with 3.77ms right? Based on these measurements, what advice could you give me for delaying the mains? By ear 10.5ms sounds better than 0ms, because the bass doesn't sound slow/late. But I'd like to get as fine a point as possible using REW. Thanks in advance.

 

[Jochie]

Here is the mdat.

 

[Keith_W]

These are much better measurements than your first effort. The SNR is much better, at least the graphs are interpretable now.

However you are only measuring your left speaker. The goal of subwoofer time alignment is to (1) remove egregious delays in your subwoofer, and (2) remove as many dips as possible in the combined measurement. This means your LR speakers need to be measured TOGETHER and your sub aligned to the combined result of both main speakers.

Anyway, let's pretend for a moment that your L speaker alone is both speakers and use the alignment tool.

Using these settings ...

I was able to achieve this result. Don't forget you can invert the polarity and play with the gains as well. It all depends on how your brain works. I like to move the slider around until I see the impulses roughly align, then I check the freq response. If there is a horrible dip, I move the slider around a bit and see if I can make the dip worse. When it is at its worst, click on "invert polarity" and voila, the dip magically disappears. If there isn't a dip, I invert the polarity and try to make the worst dip I possibly can. Then I invert the polarity back.

I have been thinking that I should write another book on subwoofer integration. But that would be a massive undertaking and the work would be a real magnum opus. To be honest, I don't have the energy to do it and I am also running out of steam with helping people on ASR.

 

[Jochie]

Thank you so much for this..! I applied your settings (mains 17.6ms delay, sub inverted, subgain -10 to taste), and it sounds really coherent. I just uploaded the measurements of both before (L+R, sub, L+R+sub) and after (L+R+sub). If you are done taking looks, I completely understand, no problem. But if you do: I notice that to my eyes, the graph with the delay and inverted sub looks a bit worse with a bigger dip at 55Hz and a slight dip around 100Hz?

 

[Keith_W]

Please be careful when taking measurements. Something strange is happening here.

This is a comparison of L+R and L+R+sub. As you would expect, the two curves are exactly the same above the subwoofer XO point.

And this is a comparison of the previous two curves with L+R+Sub, supposedly ONLY with delay to the mains AND polarity inverted. I am not sure what polarity you inverted, but look at the difference in the curves above 400Hz. The curves should be exactly the same, but they are obviously not. Are you sure you didn't invert the polarity of one speaker and not the other? As for which curve looks better, clearly the blue one is better.

Also, be careful with your reading. When I did the previous simulation, I EXPLICITLY STATED that the 17.6ms result I obtained was from an improper measurement - i.e. your left speaker only. I said you would have to redo your alignment with both speakers playing. To your credit, you did take L+R measurements this time. But that 17.6ms recommendation is out the window.

The reason I participate in this discussions is not to "give you fish" but to "teach you how to fish". I think I left enough instructions on how to use that alignment tool. Play with it a bit, generate some aligned sums, then compare them to see which gives you the best result. You have more time than we do to play with your curves, so it is likely you would get a better result.

BTW: try as I might, I couldn't get rid of that 50Hz dip. Maybe you'll have better luck. I did not try every permutation though.