Part 2 - 2.1 Stereo Calibration - what am I missing?

[Dave the Rave]

Few years back, I started the DSP journey through REW. Advice from experts here and elsewhere brought a real SQ transformation guided by REW measurement capabilities. Tried Dirac but way prefer simpler adjustments using REW.
Below is a snapshot of the measurements years back compared to now.
Key reasons for the improvements are the usual suspects recommended by audiophiles: Speaker positioning, level matching, proper measurements, DSP below 300h only, get source right with minimal latency, decent DAC/amp/speakers.
I am quite happy with the PEQ applied and only use cuts. I always found dynamics/clarity could be compromised when trying to smooth the FR towards the curve (with PEQ gains or convolution DSP).
However, i just wanted to know if there are any other techniques I can use to further improve the bass side esp tackling the nulls?
I have attached the measurements. Note, I can't go with room treatment; Most of the time the impulse graph align at zero when I place the mic accurately in the middle (hence no issue on impulse response).

Measurements 3 years back

Measurements NOW

PEQ Applied

Impulse Response

 

[Keith_W]

Look at that huge forest of early and loud reflections (ETC taken from measurement of left speaker alone). Are you using one of those mini tripods that come with the UMIK-1? This is typically what you see when you use a mini tripod and perch it on a chair, or wedge a microphone between books to take a measurement, etc. That first reflection at 0.6ms is coming from something that is about 20cm delayed from the microphone. For e.g. a wall that is 10cm away, or a sofa, etc. Your measurement needs to be taken from where your head is when you listen. Ideally you should purchase a proper mic stand, or at least duct tape your microphone to a broomstick and try to get it in a proper position. Mic tripods are cheap, my first tripod cost $50 and lasted me for years until I broke it. My new tripod is much nicer, but it still cost me only $200.

As for those dips, some of them will be real, and some of them will be artefacts from microphone positioning. You can't tell which is which from a single sweep from a fixed position.

Which Australian city are you in? I'm in Melbourne.

 

[ddude003]

You can get a nice Pro Tripod here in the States for under $100.00 USD... Try Guitar Center or Sweetwater .com...

 

[Dave the Rave]

Thx mate am from Sydney. Yep the cheapo tripod in play but taken at head position from the listening sofa.
What are the correction options with the better measurements? I will definitely get another standing tripod if that would help.

 

[Keith_W]

Hi Dave. I love Sydney. In fact I have a rental property over there in Zetland.

You do not want to correct artefacts from microphone positioning. If you mount a mic on a tripod and take a single sweep, that mic is capturing sound from a very specific point in space. If there are a lot of local reflections, there will also be a lot of comb filtering. Comb filtering causes peaks and dips in the frequency response.

Remember that your microphone has a 1/4" (6.35mm) capsule, and your ears are about 15-18cm apart. In addition, you also move around in your listening chair. Your ear "averages out" the peaks and dips in the response which change very quickly. What your ear does not average out are actual peaks and dips that do not change. For example, if your room had a massive 80Hz peak that does not change much when you move around your seat, you would surely hear that. It is the same story if you made an inappropriate EQ adjustment. The measurement would look fine (from that very specific point in space) but it would sound off.

So the real question is what to EQ, and what to leave alone. Some people will make a blanket rule that you should not EQ a single mic measurement, and you should only EQ an MMM or averaged measurements from several positions around your listening area. I think that it depends - if your frequency response does not change much over a listening area, then it's OK to equalise from a single position.

The presence of copious early and loud reflections in your ETC suggests that you are measuring (and potentially correcting) a lot of "local" phenomena. If you want to investigate this, place your microphone at the MLP (Main listening position) and do a sweep. Label it MLP0. Then move the mic 10cm, 20cm, 30cm to the left and take sweeps. Do the same for the right. You will end up with 7 measurements: MLP-30, MLP-20, MLP-10, MLP0, MLP+10, MLP+20, MLP+30. You can choose a larger area if you want to correct for a larger sweet spot.

Overlay all of them and examine the bass response closely. You will see that some of the curves "move" by quite a bit. Others may be relatively static. Then average all of them (Vector Average) and use that as basis for your EQ. The alternative is to do an MMM, but I don't like MMM's because I prefer to study the peaks and dips in detail. The advantage of the MMM is that it is quick, repeatable, and you don't have to purchase a mic tripod. Just duct tape your mic to a broomstick and you're done! The outcome is the same though, so do whichever makes more sense to you

I know you said you can't install room treatment. But ... looking at your ETC, I think you should consider it.

 

[Dave the Rave]

@Keith_W Thanks for that elaborate explanation. You did well having a foot in Zetland (prices gone through the roof).
I am not an expert at analysing all the REW components other than FR but your proposed actions are easy to understand and makes a lot of sense.
Excited to get onto the new measurements and will report back. I did manage to find a proper standing mic in the meantime as @ddude003 suggested (thx for calling it out).

 

[Dave the Rave]

G'Day,
Ran a new set of measurement over the week-end with a standing mic and MLP. Attached is the updated file and showing the mic in the listening room.
- Found that I was using the wrong cal file for the Umik mic since i have 2 !!!! Resolved that issue.
- Moved the mic as advised above. Found the R channel (ref) Impulse Response consistent unlike L channel which moved. I am assuming this is ok.
- Vector average is not the same as RMS average - not sure if this is OK as I am assuming the IR derives the Vector Average
- Applied PEQ cuts per the Target (with some manual adj). The net effect yielded BETTER results than the previous PEQ. Applied < 300h only.
So one good win through that long process. Happy to get any other feedbacks.

 

[Keith_W]

Let me show you something.

These are the impulse responses for all your left sided measurements.

If you select them and click "Align IR start", this is the result. The reason why this matters:

Compare the vector average you obtained without aligning IR start (green) with mine (purple). In this case there is very little difference between them below about 500Hz or so. If you were planning to DSP the high frequencies above 500Hz then this kind of summation error matters.

I was telling you about the difference between "real" dips in the response (room related) vs. "local" dips (caused by proximity of the measurement mic to furniture). This overlay of all your left sided measurements shows you the difference between the two quite nicely. If you checked over a wider area, you would observe bigger differences. Not that you need to, if this is where your head is positioned when you listen, that's all you need to correct for.

Anyway it appears that in your case, your measurement does not seem to change much over a +/- 30cm listening area. So you can go ahead and correct it with PEQ.

 

[ddude003]

Hi @Dave the Rave... I hope you don't mind if I am a little picky... I have looked over your room photo and have some comments... First, I corrected the photo tilt a bit to align with the back wall... Then I noticed that the center of the room seemed to relate via the back window... I try and show this in the photo below...
So, speaker placement as well as mic placement are pretty important to get speaker time alinement and mic perfectly centered...
And your left speaker seems to be toed in more than the right... Maybe its just the photo?
I love those hard wood floors... And they play havoc in a small room... I would suggest some/more physical room correction to complement your digital room correction... Some Bass absorbers and some wide band absorbers would be of great benefit...

 

[Dave the Rave]

Hello, Thanks heaps for the feedback. Appreciate those much better.
I used the following DSP steps:
1. IR time aligned to generate vector averages. Generated LR (Vector average) from these R & L channel averages.
2. Applied PEQ at 1.0 slopes above and below 200h at 66db with Var graph. Applied to L, R and LR
3. Vector averaged EQ-L, EQ-R and compared with LR Average in step 1.
4. Found out I was cutting too much from PEQing L & R channels for some frequencies. Adjusted L & R PEQs accordingly to align closer to EQ-LR
Net result - BEST PEQ effect in my system so far. Comparing the same tracks, bass is firm and taunt. Mid and High sound more open.

Understand I have more room for improvement for the dips with some Bass absorbers. If these can be placed with WAF considerations, will look into it (and re-measure with the proper center placement).
Would it be of any benefit placing the absorbers behind the speakers (hidden treatment = less WAF negociation = less spend on shoes shopping for misus)?

 

[Keith_W]

You don't need bass absorbers. I'll show you why.

First, this is the vector average of all your measurements after aligning IR start for left (red) and right (green). I then Vector summed both of them to get an idea of what your overall bass response would look like (purple). I staggered the purple curve for clarity. And bear in mind this is without ANY DSP to improve the response.

I then applied ERB smoothing to the vector sum. Here it is in red compared to the original vector sum in faint purple. Notice that all the dips disappear.

If you want to understand what ERB smoothing is, watch this video by Amir from ASR. Start from 3:55.

Basically, your hearing is nowhere as "hi res" as your microphone. Yes, those dips look ugly on a graph, but you won't hear them. The exception is if you have very wide and deep dips, which you don't. The reason we correct graphs is so that we can correct beyond the limits of our hearing.

If you REALLY wanted to do something about those bass dips, room treatment is the wrong approach. Bass absorbers need to be 1/8 as thick (at a minimum) of the longest wavelength you wish to absorb, and its effectiveness depends on very specific placement and how much surface area you treat. At that thickness it is really intrusive and will have all sorts of undesirable side effects - like killing off all your high frequencies. I can guarantee that some thin foam behind your speakers will do next to nothing.

If obtaining a super-smooth bass response is your goal, you would be better off with multiple subwoofers and careful DSP.

 

[Dave the Rave]

@Keith_W Thanks for the explanation and the video is genuinely instructive. This definitely settled down the FOMO aspect to some extent.
Obviously the curiosity took over and I started reviewing the ERB concept.
Since a long time, I dialed down the mid/high knobs of my speaker (Yahama NS-1000x) and never fiddled with them again - reason being I always got less than satisfactory results when correcting above 300h.
However, after listening Amir's take of how humans actually hear sound (as you pointed out as well that the microphone is just a high-def fixed listener), I reset my speakers to a neutral state (ie no cuts in mid/high) and did some basic measurements.
I applied PEQ as shown in the picture below i.e
- I kept PEQ applied up to 300h as per my earlier post
- For FR above 300h, I used ERB smoothing and applied a single minimalist PEQ with a wide Q or Width of 0.7 - as Amir highlighted no point correcting towards a fully smooth target curve.
As the new measurements shows, sound is more lush and so far I don't hear weirdness (compared to my previous experience in applying PEQ across all frequencies).
Will keep listening over a longer period and see if applying PEQ across all frequency range genuinely works.

 

[Keith_W]

Those Yamaha NS-1000's are notorious for having a wonky frequency response. This can be fixed, but correcting an in-room measurement is the wrong way to do it. You need to obtain the quasi-anechoic minimum-phase response of your speaker.

Take your speakers outside and elevate them as much as you can. Try to ensure there are no nearby reflective objects (fences, walls, etc). Place your mic at least 2 baffle widths away, or about 1m. Take a sweep, then examine the ETC or impulse response for reflections. If you see a reflection at (say) 10ms, this means that all frequencies with a period longer than 10ms (100Hz) have been contaminated by reflections. Gate these out using REW's MTW function, then make a minimum-phase copy. Use that as the basis for your correction.

Anything below 100Hz in this scenario can be corrected together with the room.

 

[Dave the Rave]

Thanks Keith. I will get my head around the MTW function and try the suggested option.
Just to clarify, how do you use the IR/ETC to determine the reflection (my basic understanding below) -
Is the first reflection happening at 0.8m i.e 27cm (.34m per 1m) - Second bigger reflection at 2.4m i.e 82cm?
Also, any harm using the min phase copy as per the current measurement?

 

[Keith_W]

In the screen shot you posted, these are the reflections:

You look for the same thing in the ETC. Switch the vertical scale from % to dBFS and look for spikes. Measure the time-lag between the spikes and the initial impulse. This will tell you:

- the period T of the longest wavelength which is uncontaminated by reflections. In your case this is about 0.6ms. f = 1000/T (T in ms) this is 1667Hz.
- the extra distance the reflection had to travel to reach the microphone. d = c * t/100 (d = distance in m, c = speed of sound 343m/s, t = time in milliseconds). For 0.6ms, this is 200mm. My guess is that it's a reflection from the sofa.

 

[ddude003]

Of course there are a lot of reflections... Its a bare room for gods sakes...

You might take a look at http://arqen.com/bass-traps-101/placement-guide/ for some other guidance...

 

[Dave the Rave]

@ddude003 Did some reading on the placement/treatment guide per your post. I will be looking for possible treatment over the long term. However, I did revisit the speaker placement and ultimately, I moved the speakers away from the backwall. I previously had them close to the backwall as Yamaha users seem to better like them as such with these speakers (note the NS1000x came after the NS 2000 and supposedly improved i.e much better than the Ns1000).

@Keith_W My understanding of IR/ETC/GD/Waterfall improved thanks to your post and after reading your well written explanation in another forum of how to measure and common mistakes people make using REW - fantastic resource to layman like me.
I did some quick measurements after all the readings and speaker placement as below.

1. Measured L, R and L+R without any DSP at listening level

2. Used the speakers mid/high knob to correct some frequencies to ensure Left and Right channels FR are closely aligned
Ended up removing some Mid to the Left (-2db) and some treble to the right (-2db)
Note, I have previously used Balance/Gain in lieu of Mid/High fiddling to bring both channels' FR closer but I always ended up towards neutral balance for better transcient.

3. Measured combined L+R after step 2 with the intention is to use it for DSP
I did a lot of reading on merits of Eq'ing by channel or combined L+R and many people tend to suggest L+R Eq maintains tonality, phase and centre balance

4. Tried Var smoothing based PEQ by combined or per channel with some good results but something always start missing when listening over longer time (as these generate many PEQ some with high Qs)

5. The revelation came when I applied two simple PEQs with high Q using a mix of smoothing i.e ERB (up to 500h) and Var (above 500h)
This sounds the best to my ears so far as it maintains dynamics, gives tighter bass, controlled mid and highs drizzled with proper flesh. Some of these attributes were lacking with my previous DSP experience.

FYI the measurements are quick ones hence the IR may be all over the place. I am sure I can fix those by multiple measurement and proper alignment. However, I am not sure how much these would change the PEQ values as only wide Qs are being used. I understand there are reflections given an open room but I don't know if there are solutions to such issues other than room treatment.

 

[ddude003]

Perfect or near perfect symmetrical speaker placement in a room like yours is very important... And here is a link to one of the finest acoustic treatment companies... GIK Acoustics https://www.gikacoustics.com... I am not suggesting you buy their products... What I am suggesting is that you can take a look at their product selection and learn some basics... They can also look at your room acoustics and make suggestions on how to deal with any issues... You can also make your own bass and wide band absorbers... I have made my own and bought two with my own selection of art from GIK... This gave me a good starting basis to which I add DSP Convolution+FIR filters and then layer an Old Skool Passive Parametric EQ to taste... Here is a peek at my kit... https://www.avnirvana.com/threads/a-small-a-v-entertainment-room.5662/

 

[Dave the Rave]

Pheeew mate what a nice setup - will die to hear how it sounds like. Or better not as it might spur a divorce to get my man-cave back!!
Heard good things about GIK Accoustics in the local stereonet forum. Time to venture out a bit.
Interesting you use Convolution - I always found them cleanly delivering across all frequencies but they tend to impact dynamics when listening some genre hence i moved to PEQ only. I have a MiniDSP SHD which allows proper sub crossover and Dirac DSP+PEQ and again i tend to go back to the default sound of the speakers. Hence the last iteration of minimal PEQ actually sounds best so far.
Obviously I keep getting improvement with all suggestions here so will keep an open mind.
My Setup: Rpi5 host (Streamling/Local file) = DIrettaRendererUpnp ==>> Pi2AES (Target) ==>> Yggdrasil DAC =Balanced==>>Yamaha AS2100 Amp ==> Yamaha NS1000x speakers / JBL Sub

 

[ddude003]

Thank you... Everyones kit, room, ears and brain are unique... I feel that you must trust your own ears as you move forward tweaking your kit and room until you reach AVNirvana... And it looks like you are getting there my fren...

You might look online for a version of the Master Handbook of Acoustics by F. Alton Everest & Ken C. Pohlmann... For a sip from the fire hydrant and watch

if you are looking at Convolution+FIR... ...

 

[Keith_W]

Interesting you use Convolution - I always found them cleanly delivering across all frequencies but they tend to impact dynamics when listening some genre hence i moved to PEQ only.

FIR does not impact dynamics. Like all DSP, it sacrifices volume for linearity. Badly implemented DSP (i.e. ANY DSP with aggressive volume cuts) limits maximum volume and can make a system sound tame. You can also do all sorts of things in FIR that you can't do with IIR, so the temptation is to try to make the measurement look as perfect as possible. This can come with all sorts of nasty side effects if one is not careful.

DSP is just a tool, and the more powerful the tool is, the more possibilities there are for screwing it up.

Anyway I just came across a nice article by Archimago about small room acoustics with some discussion about room treatment: here.

I see you are in Australia. Which city? I am in Melbourne.

 

[Dave the Rave]

In Sydney unfortunately. Sounds like i should give Convolution another go which I am keen to do next after a proper measurement set.
Spot on about the observation as I indeed tried getting a perfectly flat response across the full FR in the past (measurements were not great).
I was using the inversion method by Obsessive Compulsive Audiophile and it yielded great results which I used for about a year. However, once I mistakenly disabled DSP and heard what I was missing - that's where I starting drifting to minimal DSP.

Any suggestion in regards to the below:
1. Any other method of creating FIR using REW ?
2. Is it recommended I use DSP below 300h per my FR curve (see the quick measurements above)?
Heard some ppl say it should be applied where the speaker crosses over - mine is at 500h and 6000h. If so, are there any guide for partial Convo creation technique?
3. Will consider any suggestion on the target curve which I can adjust to my personal preference ?

Thanks

 

[ddude003]

If I didn't know any better, it seems like Archimago is suggesting in that article the same thing that I, and others, had suggested for years now... Get a start on your small room correction with some physical bass traps and wide band absorbers... Maybe even some diffusers... Then add some DSP...

REW is a measurement tool and not directly a FIR filter creator... The author has even said so... Can you? Yes... Are they the best and/or the SOTA there? Maybe not... Sure, I have tried AOC's methods early on... Did you use rePhrase too? Better than nothing... And have you tried AOC's latest? For a novice here are a few free to try...

https://github.com/ObsessiveCompulsiveAudiophile/GSonic/releases/tag/v1.0.19 the latest from AOC (serko70) here...
and https://github.com/xPoiler/XPDRC Another AVNirvana member...
and https://github.com/VilhoValittu/DecayCore/releases/tag/v1.0.0 Another interesting solution...

I am sure Keith_W knows of a few more...

 

[Keith_W]

1. Creating FIR with REW is possible if you use it with rePhase. It's not the smoothest or the easiest experience and it will surely make you want to tear your hair out, particularly if you are trying to correct a multiway speaker and generating filters for each driver. You REALLY need to know what you are doing, and I DO NOT recommend it.

There are free options, e.g. LinFIR and GSonic. I have not used either, so I do not know their capabilities. They are free to try, so you could certainly have a look. Otherwise, you could use Acourate or Audiolense. I use Acourate, IMO it's the most powerful and most flexible FIR design tool on the market. It's rather teutonic in its design, e.g. the author insists that you make your own timing signal and examine it manually. It really forces you to think about what you are doing and whether you want to do it or not, a bit like REW.

2. Re: DSP below 300Hz. You can DSP the entire frequency range, provided you are able to take the correct measurements and you know what you are looking at. Below 200-300Hz (depending on your room size) you correct speakers together with the room - "room correction". Above the transition zone, the speaker dictates the frequency response. So you need a quasi-anechoic measurement of your speaker. This kind of measurement may not be easy to take (depending on your speaker), and once again you need to know what you are doing. You have to ask yourself whether you should attempt it or whether you are better off leaving it alone.

3. Re: target curves. That's an entire debate which I don't want to get into. There are huge threads on ASR debating the merits of target curves, including Floyd Toole himself who said that it's a target result and not something you force your speaker to comply. All i'll say is: make up your own mind.

@ddude003 AOC is an American politician. @serko70 is OCA.

Re: bass traps. If Archimago said that, then I don't agree with him. The problem is that velocity absorbers / porous absorbers are physically intrusive and too broadband. Pressure absorbers (Helmholtz resonators and membrane absorbers) are difficult to tune, physically intrusive, and too narrow band. In fact it looks like a single PEQ. The only bass trap I would consider are VPR's (Verbundplatten Resonators), they are more broadband, less intrusive, and can be tuned by adding mass. I should go back and look at his article again.

 

[Dave the Rave]

I did use RePhase as it was an optional task in Serko70 methodology - it made a good impact but was painful and I did not know what to achieve exactly. Will go through the reading list and tools suggested (albeit will take time to absorb). Thanks for the suggestions.