Keith_W DSP system

[Keith_W]

Anyway, it is the weekend and I can play with my DSP again. I decided to redo everything from the ground up, because why not. I redid the time alignment via sine wave convolution method again, and was getting very strange results comparing left sub and right sub. So I compared them to each other:

Red = Left sub convolved with 50Hz sinewave tone
Green = Right sub convolved with 50Hz sinewave tone

It looks as if the subs are out of phase with each other. However, the initial deflection of both subwoofers is in the same direction. I was a bit baffled by this point. So I decided to take a mono recording (i.e. left + right playing together) to look for comb filtering.

Red/Green = uncorrected left and right sub + speaker response.
Brown = mono recording of left/right sub + speaker
Blue = mono recording of left/right sub + speaker, but with polarity inverted in the right sub.

As you can see, there is comb filtering in the uncorrected response. This disappears when the polarity of the right sub is inverted.

At this point I realized that the entire correction process was made with the polarity of one sub inverted. This is easy to do if you simply let Acourate handle both L/R without turning on your brain, those macros do it automatically and don't complain. So what does the mono recording of the corrected curve (with subwoofer polarity inadvertently reversed) look like? It looks like this:

Acourate seems to do a remarkable job of coping with the inverted polarity of one sub relative to the other. I must have inadvertently inverted the polarity of the subwoofer after moving the system. This is easy to do - there are two unlabelled Speakon connectors on my DIY subs, and if you connect them the wrong way, it will invert the polarity.

Anyway, the point of this post is: we rarely do mono recordings in our verification measurements. The lesson for me is to do them more often, because you can find things like inadvertent polarity inversion in one pair of drivers!

 

[Keith_W]

Well, after discussion with @OCA he pointed me towards this video:

... which made sense to me. Align the phase of the subwoofer and the woofer. So now I have a 4th time alignment method. I will describe it here for Acourate users.

METHOD 4: PHASE ALIGNMENT OF SUBWOOFER TO WOOFER

1. Take a measurement of subwoofer+tweeter and woofer+tweeter using the filters created in Step 3 of this post. You will obtain the above result.

2. Switch to the phase view and select "unwrap phase". The zoomed out view looks like the above.

3. Now we zoom in to the frequency of interest. Above shows 10Hz - 100Hz, note the phase angle on the left has been adjusted to 0 to -100 radians.

4. I want to time align the subs and woofers at 50Hz, so an extreme zoom in will allow you to read the angles. Here, it is -47.547 radians for the subwoofer at 49.805Hz. The woofer is -44.777 radians @ 49.805Hz. The discrepancy between the two is 4.455 radians.

5. Now we do some math to convert the discrepancy into time, and then into samples.
- First convert radians to degrees: radians * 180 / pi = (4.455 * 180)/3.141 = 255.252 deg.
- Convert degrees into time in milliseconds: (degrees*1000) / (360*frequency) = 255.252/(360*49.805) = 14.2362ms.
- Convert milliseconds into samples: (sample rate / 1000) * time in ms = 48*14.2362 = 683 samples

6. Rotate the subwoofer by -683 samples and we obtain the above result. We can see that the subwoofer is time aligned to the woofer at 50Hz, but nowhere else. I am starting to realize that this is the nature of the beast ... the behaviour of bass frequencies can never be perfectly aligned, even with FIR filters.

7. Now we check the alignment using the sinewave convolution method described in this post. Rotate the sub by -683 samples and we obtain the above result. We can see that the initial impulses (left of the graph) are not aligned, but the steady state is aligned.

8. Now that we know that the method works, we go back to the phase view and nudge the subwoofer and woofer closer and closer together until they are mostly aligned at as many frequencies as possible. Here, I favoured closer alignment between 40-50Hz at the expense of lower frequencies, because those are less audible. Total rotation = -763 samples. You could simply start nudging at step 3 of this process (and skip the rest of the steps and the math), but I included it to prove that the method works.

... Everything isn't as rosy as the picture above shows however. If I cross-correlate it with the time alignment by step response method, you can see that it is completely off:

I am still trying to wrap my head around whether this is a valid method or not. The problem with ALL my subwoofer time alignment methods is that very few of them cross-correlate. This is because some of them align the initial pulse, but not the steady state (e.g. the step response method shown above). As Uli said to me, "subwoofers are always a headache". (He hasn't seen this method yet, I wonder what he thinks).

To sum up, I have learnt that:
1. The initial impulse and the steady state of the subwoofer are two different things. Alignment for one might mess up the other. However, I favour alignment of the steady state because the initial impulse is short. This involves sacrificing a perfect looking step response.
2. The time delay varies according to the frequency of interest and therefore the phase; it is therefore impossible to time align the subwoofer so that all wavelengths are aligned.

The advantages of this method are:
1. Allows you to look at every wavelength of interest and align them to the best of your ability,
2. It is simple to do

 

[dualazmak]

Well, after discussion with @OCA he pointed me towards this video:

... which made sense to me. Align the phase of the subwoofer and the woofer. So now I have a 4th time alignment method. I will describe it here for Acourate users.

METHOD 4: PHASE ALIGNMENT OF SUBWOOFER TO WOOFER

1. Take a measurement of subwoofer+tweeter and woofer+tweeter using the filters created in Step 3 of this post. You will obtain the above result.

View attachment 353874

2. Switch to the phase view and select "unwrap phase". The zoomed out view looks like the above.

View attachment 353884

3. Now we zoom in to the frequency of interest. Above shows 10Hz - 100Hz, note the phase angle on the left has been adjusted to 0 to -100 radians.

View attachment 353892

4. I want to time align the subs and woofers at 50Hz, so an extreme zoom in will allow you to read the angles. Here, it is -47.547 radians for the subwoofer at 49.805Hz. The woofer is -44.777 radians @ 49.805Hz. The discrepancy between the two is 4.455 radians.

5. Now we do some math to convert the discrepancy into time, and then into samples.
- First convert radians to degrees: radians * 180 / pi = (4.455 * 180)/3.141 = 255.252 deg.
- Convert degrees into time in milliseconds: (degrees*1000) / (360*frequency) = 255.252/(360*49.805) = 14.2362ms.
- Convert milliseconds into samples: (sample rate / 1000) * time in ms = 48*14.2362 = 683 samples

View attachment 353887

6. Rotate the subwoofer by -683 samples and we obtain the above result. We can see that the subwoofer is time aligned to the woofer at 50Hz, but nowhere else. I am starting to realize that this is the nature of the beast ... the behaviour of bass frequencies can never be perfectly aligned, even with FIR filters.

View attachment 353888

7. Now we check the alignment using the sinewave convolution method described in this post. Rotate the sub by -683 samples and we obtain the above result. We can see that the initial impulses (left of the graph) are not aligned, but the steady state is aligned.

View attachment 353890

8. Now that we know that the method works, we go back to the phase view and nudge the subwoofer and woofer closer and closer together until they are mostly aligned at as many frequencies as possible. Here, I favoured closer alignment between 40-50Hz at the expense of lower frequencies, because those are less audible. Total rotation = -763 samples. You could simply start nudging at step 3 of this process (and skip the rest of the steps and the math), but I included it to prove that the method works.

... Everything isn't as rosy as the picture above shows however. If I cross-correlate it with the time alignment by step response method, you can see that it is completely off:

View attachment 353899

I am still trying to wrap my head around whether this is a valid method or not. The problem with ALL my subwoofer time alignment methods is that very few of them cross-correlate. This is because some of them align the initial pulse, but not the steady state (e.g. the step response method shown above). As Uli said to me, "subwoofers are always a headache". (He hasn't seen this method yet, I wonder what he thinks).

Just wonderful! What a lot of really interesting and invaluable experiments and results you are sharing with us!!

For optimal time/phase alignment between the SP drivers, especially between subwoofers and woofers, which would be better by "actual sine tone-burst wave matching with careful timeshifts/group-delays" or by "simple step response", at/around optimal XO Fq??

Myself and @DrCWO have rather intensively discussed this theme around here and somewhile thereafter (until to here, here and his educational/theoretical post here); throughout our discussions there, @DrCWO has been always sticking to his own "step response" approach, on the other hand I was sticking to my primitive "sine tone-burst wave shape matching" method (ref. here, here and here) (as I have never tried "step response" approach, though.)

Since I could have established "my optimal" XO Fq and time/phase alignment through my primitive "sine tone-burst wave shape matching", I do/did not need to have conclusion for which of the two approaches would be superior to the other, even though I well understood what he would be (theoretically?) insisting. (And, vice versa, it looks he never tested the "sine tone-burst wave shape matching".)

Your present on-going intensive experiments and data thereof, as well as the comment "subwoofers are always a headache" of Uli, would seem to tell me that we have no concrete answer for which is better/best, but we may better to, or need to, have try-and-error processes using both (or several) approaches for our own SP audio gears (drivers) in our individual acoustic room environment to find our own optimal XO Fq and filter/phase/time-alignment configurations.

Yes, this issue is always a kind of headache in DSP multi-SP-driver multi-amplifier audio system, but once we would establish our own optimal configuration, we would definitely be able to enjoy new wonderful horizons and perspectives of our audio listening experiences (as I myself do enjoy now).

 

[DHT 845]

I just noticed this thread and I must read it all since I was a fan of acapella speakers since I heard Violon 1 with Gryphon Antileon and Wadia 860 like 30 years ago or so...

 

[Keith_W]

Another concern Dr. Uli had was that the system was not symmetrical. In particular, the right subwoofer is blocked from the MLP by a sofa:

This is the view from the microphone placed at the MLP:

I wanted to see if the sofa really does screw up the frequency response, phase, etc.

So I removed it and took some measurements without the sofa, and with the sofa. The microphone was not moved between all the measurements. Although I did check the left subwoofer as well, the subwoofer of interest is the right. So I will only be showing measurements of the right sub.

First the frequency response - brown is with the sofa, blue is without. Both exhibit a dip at 40Hz (as will be seen in the GD plot below). The magnitude of the dip is slightly less without the sofa.

Group delay. Both broadly have the same shape, including the spike at 40Hz, but the amplitude of the spike seems a little lower without the sofa.

Unwrapped phase. Again there is almost no difference. I would have thought that reflections from the sofa would really mess this up, but it does not seem to make any difference at all.

Conclusion: Surprisingly the sofa does not seem to make much of a difference despite obstructing line of sight to the subwoofer. Now I can have peace of mind having the sofa where the wife wants it, knowing that it does not mess with the performance of the sub too much.

 

[Keith_W]

Just wonderful! What a lot of really interesting and invaluable experiments and results you are sharing with us!!

For optimal time/phase alignment between the SP drivers, especially between subwoofers and woofers, which would be better by "actual sine tone-burst wave matching with careful timeshifts/group-delays" or by "simple step response", at/around optimal XO Fq??

Thanks for your response. I am not familiar with those methods yet, but I will take a look at them and see if I can use them. Not all methods can be used with Acourate, e.g. @OCA described another method to me using alignment of group delay ... but Acourate automatically aligns the time to the tweeter. There is probably a way to get around that, but it is a bit too much for my small brain to handle.

I just noticed this thread and I must read it all since I was a fan of acapella speakers since I heard Violon 1 with Gryphon Antileon and Wadia 860 like 30 years ago or so...

Thank you, but these are much superior to the standard Violons Standard Violons have bass problems and no DSP. These things will absolutely murder a standard Violon any day of the week. They do make for a rather entertaining speaker to DSP though. They have just enough problems to have fun fixing, but not so many that fixing it is insurmountable. In any case, the Violons do not give me too many headaches. It is the subwoofers which are giving me no end of problems, but that is not Acapella's fault. I designed the subs, so it is all on me.

 

[OCA]

the behaviour of bass frequencies can never be perfectly aligned, even with FIR filters.

You could possibly almost perfectly sync their phase response AT THE MLP with a number of allpass filters some of which would have to be time-inverted and with high Q parameters if you DON'T MIND hearing echos from the future!

 

[OCA]

Thanks for your response. I am not familiar with those methods yet, but I will take a look at them and see if I can use them. Not all methods can be used with Acourate, e.g. @OCA described another method to me using alignment of group delay ... but Acourate automatically aligns the time to the tweeter. There is probably a way to get around that, but it is a bit too much for my small brain to handle.



Thank you, but these are much superior to the standard Violons Standard Violons have bass problems and no DSP. These things will absolutely murder a standard Violon any day of the week. They do make for a rather entertaining speaker to DSP though. They have just enough problems to have fun fixing, but not so many that fixing it is insurmountable. In any case, the Violons do not give me too many headaches. It is the subwoofers which are giving me no end of problems, but that is not Acapella's fault. I designed the subs, so it is all on me.

If I remember from Mitch's video correctly, Acourate has a "magic" tool which clears excess phase "differences" between speakers at points you select. Why don't you use that to align your woofer and subwoofer?

 

[Keith_W]

Hello @OCA. You are talking about Acourate's Interchannel Phase Alignment (ICPA). This is the ICPA in action:

As you can see, it looks for differences in the left and right channels in the group delay plot. You are not supposed to normalize outside the Q limits (indicated by the orange and purple lines), or spikes in the GD that occur symmetrically in both channels. There are a few other guidelines that Uli published in the white paper.

It is not a time alignment tool.

 

[OCA]

It is not a time alignment tool.

Looks like a great tool to sync your woofer with the subwoofer to me

 

[3ll3d00d]

I did this in the past to adjust the phase response of one sub to match the phase response of the other sub, seems similar to what you want? You basically take the excess phase from one and convolve it with the xo for the other.

 

[DHT 845]

Thank you, but these are much superior to the standard Violons Standard Violons have bass problems and no DSP. These things will absolutely murder a standard Violon any day of the week. They do make for a rather entertaining speaker to DSP though. They have just enough problems to have fun fixing, but not so many that fixing it is insurmountable. In any case, the Violons do not give me too many headaches. It is the subwoofers which are giving me no end of problems, but that is not Acapella's fault. I designed the subs, so it is all on me.

From what I remember bass was punchy but slightly overwhelming the mids, adding kind of muddiness (a bit). But I liked the mids and ion tweeter. I even wanted to buy just tweeters (they were available in separate boxes). But eventually I went into true ribon road and planar magnetics like BG. Now I use traditional drivers. But next week I will be implementing new midrange concept with separane enclosures based on 2x 3inch magnetic fluid drivers (800-5000Hz) Above 5000 is 22mm wavecor without ferrofluid.

 

[Keith_W]

Looks like a great tool to sync your woofer with the subwoofer to me

Yeah, not the way it has been implemented in Acourate, I think. The ICPA option is offered as the second last step in the room correction macros, only before final generation of filters. It won't even run if you haven't gone through the previous steps (i.e. making a minphase version of your sweep, creating a target curve, convolving the target curve, etc). Even before you wade into the room correction macros, you should try to make your speaker "as perfect as possible". This means driver linearization, time alignment, volume matching, etc.

 

[Keith_W]

Actually, my brain has gone missing in the past few days. I was looking so closely at aligning the subwoofer to the woofer, and worrying about aligning all the frequencies ... that I forgot that I cross my sub over to the woofer at 50Hz. The bass frequencies don't need to be aligned - only the crossover point and maybe half an octave above/below that because the XO is pretty steep. Stupid, stupid!

 

[dualazmak]

I too now cross L&R subwoofers (YAMAHA YST-SW1000, 30 cm ported YST system ref. here, 48 kg each) and woofers (YAMAHA JA-3058, 30 cm, in sealed NS-1000 cabinet) at 50 Hz, after intensive subjective and objective assessment of their transient behavior around 30 Hz - 70 Hz zone!

More precisely, for subwoofers I have DSP Low-Pass (high-cut) Linkwitz-Riley -12 dB/Oct at 50 Hz (polarity inverted) in DSP EKIO plus -24 dB/Oct LP at 55 Hz in YST-SW1000 itself, for woofers I have DSP High-Pass (low-cut) Linkwitz-Riley -12 dB/Oct at 45 Hz with 16.0 msec group-delay time-aligning to subwoofers.

The DSP EKIO's "digital domain theoretical SPL and phase representations" show some (more or less) complexities for which I do not care, and I do trust my ears and brain as well as microphone-recording measurements at my listening position while all the L&R SP drivers are singing together (using well QC-ed flat white noise). Of course the relative gains between the two can be flexibly remotely controlled by DSP EKIO and/or YST-SW1000's gain dial plus A-S3000 integrated amplifier driving woofers; all of these are very similar to what you are intensively doing right now. (YST-SW1000 also has on-the-fly phase/polarity-inversion control can be switched by remote while sitting at my listening position.)

Throughout my intensive efforts in DSP (and in-analog-level) XO configurations during the past several (almost five) years, I learned well that the simpler the DSP manipulations, the better the sound quality in total (the more natural sound can be heard). "In DSP, the simpler the better, the simplest the best" would be my policy now, even though I have carefully decided my present DSP EKIO's XO/Gain/EQ/Group-Delay configurations (ref. here).

At least in my case, I am always rather hesitant to use any "advanced complicated/sophisticated DSP software tool" which includes many black-box type (magic?) internal procedures beyond my naive/primitive understandings since I often experienced more-or-less "lack of sound naturality" (sorry for this subjective wording) when I intensively tested such advanced DSP tools.

I am very much looking forward to knowing your general policy/approach in DSP and/or analog XO after you have done all of your intensive efforts shared in this invaluable thread.

 

[GaryY]

More and more, I'm convinced hifi is the hobby of genius. I'm not....

 

[5-pot-fan]

Actually, my brain has gone missing in the past few days. I was looking so closely at aligning the subwoofer to the woofer, and worrying about aligning all the frequencies ... that I forgot that I cross my sub over to the woofer at 50Hz. The bass frequencies don't need to be aligned - only the crossover point and maybe half an octave above/below that because the XO is pretty steep. Stupid, stupid!

Good report! I was not reading the previous posts with full concentration and was wondering (like you must have been) how this problem could be overcome- then you provided the answer...
When you have the energy please let us know the results after re-matching at the subwoofer/woofer XO.
Also good to see that a sofa does not have a great bearing on the lower frequencies in real life.

 

[Keith_W]

I have managed to time align the subwoofers now. I will say a few things about subwoofer time alignment which will hopefully help someone out there. I took a leaf out of Dr. Uli's book, and decided that I will use multiple methods to cross-check subwoofer time alignment. I have gone over the procedure so many times and spent so much time staring at the curves that I could write a treatise about it, but I suspect that even the eyes of DSP nerds on ASR will glaze over in utter boredom. So here is the abbreviated version.

Keith's rules of subwoofer time alignment:

1. You can align for the initial impulse OR the steady state, but not for both. You have to choose. This law is immutable, it is due to physical behaviour of your drivers AND the behaviour of bass in the room (it may appear to change phase due to reflections before it arrives at your mic). There is no way to DSP around it.

2. If you align for the steady state, only one frequency can ever be time aligned, and all other frequencies will diverge from perfect time alignment. You may need to employ additional strategies to achieve better time alignment.

3. Every time you do a sweep, the result will be different. I have measured variances between 0.02 - 0.1ms. This is normal, don't get too hung up about it. You are unlikely to hear a subwoofer misalignment of 0.1ms.

4. Different filters will produce different delays, so it is important to repeat the time alignment procedure every time you design a new filter for the sub, or for the woofer.

5. The importance of a good cup of coffee can not be overstated.

So why should you time align for the initial impulse? Answer: if you are high passing the main speakers. In this case, the alignment of lower frequencies of the main speaker does not matter because the sub will be handling those frequencies. Why should you time align for the steady state? Answer: if you are using both your mains and subs to produce bass.

As proof of my first statement, that it is impossible to achieve time alignment for both the initial impulse and the steady state, I offer this graph as proof:

This is a sweep of the subwoofer and the woofer, convolved with a 50Hz sinewave to better show the time behaviour of the drivers. You can see that I have rotated the subwoofer to align with the woofer at the initial impulse, but doing this messes up the steady state. Almost as if the subwoofer has inverted polarity! But not so, look closely at the initial impulse and you will see the shape of the deflection of both drivers is the same.

If we align for the steady state at 50Hz, the initial impulse is no longer aligned, as we can see here:

However, there is not much good in doing this. This is because the phase rotates across the frequency range of both the subwoofer and the woofer. What this means is that only one frequency can ever be time aligned, and all other frequencies will diverge from perfect time alignment. I will show this graph as proof:

This is the unwrapped measured phase of both the sub and the woofer, zoomed in the frequency band of interest (20Hz - 100Hz) and the vertical scale adjusted. Using the above method (sinewave convolution), I rotated the sub and woofer by the values I derived. As you can see, perfect alignment at 50Hz. But nowhere else. Some of those delays are almost 10 radians apart. I posted the calculation to convert radians to time here - we are talking delays of up to 30ms! Now 30ms may not matter if it was in isolation, but if you have two drivers playing the same frequency 30ms apart it is bound to cause problems.

It MAY be possible to extract excess phase and use the min phase version to design some kind of all pass filter that will straighten the phase of both the sub and the woofer to get them to align a bit better. But since I don't need to do it because I am high passing the main speakers, I did not try this experiment.

This is why I like Acourate - it is not a "DSP black box" where measurement goes in, and result comes out. You have no idea what the "black box" did and whether the result is correct. I know that some software packages don't even let you change what the software has decided, nor do you even get the tools to check yourself. Acourate forces you to think and make decisions, and you learn so much in the process.

(EDIT) after all that I forgot to post a step response to show proper driver time alignment. So here it is:

What does a time aligned system sound like? Well, it is easier to describe what a non-time aligned system sounds like. If you have the ability, you should deliberately mess up the time alignment of your system. You will hear smearing and a loss of clarity. The image is unstable and seems to shift depending on pitch (the worst offenders are instruments with a wide frequency range like a grand piano). Bass sounds "slow" and laggy. Dynamics are affected, and the system sounds lazy and "constricted". With time alignment, there is incredible transparency, slam, and punch.

 

[Keith_W]

I am very much looking forward to knowing your general policy/approach in DSP and/or analog XO after you have done all of your intensive efforts shared in this invaluable thread.

I am sorry to disappoint you, but I don't have one. About the only firm belief I have is in the power of DSP. As far as DSP goes, I am up for anything. After all, it doesn't cost me any money to read about something and wonder whether I can do that with the tools that I already own.

More and more, I'm convinced hifi is the hobby of genius. I'm not....

I am far from a genius. I do not invent things, or discover things. I only follow. And sometimes I don't even follow, I ignore advice given by much smarter people than me if they haven't given me a good reason why my idea is stupid (in which case, the only way to find out is to try it myself and discover why it is stupid). It costs me nothing to try, that is the beauty of DSP Bad ideas can always be discarded, DSP can always be redone.

 

[gnarly]

However, there is not much good in doing this. This is because the phase rotates across the frequency range of both the subwoofer and the woofer. What this means is that only one frequency can ever be time aligned, and all other frequencies will diverge from perfect time alignment. I will show this graph as proof:

Hi, you may find a snip from a Smaart manual useful. (Skip the linear-phase prose at the beginning and jump down to 'Comparing Phase Traces.')

It MAY be possible to extract excess phase and use the min phase version to design some kind of all pass filter that will straighten the phase of both the sub and the woofer to get them to align a bit better. But since I don't need to do it because I am high passing the main speakers, I did not try this experiment.

What is done for live-sound sub to main alignments, is to choose a sub low-pass and a main high-pass, such that when time alignment is correct, the phase traces completely overlay each other for as wide a frequency range as possible.

(This is the same process as for putting together multi-way sections in a speaker.....get the phase traces to overlay.
Via choices in xover types, orders, and frequencies, EQs, delay, etc.)

So yes, is is possible to get the phase of a sub and main to align over a wide range of frequencies. Via the time honored phase trace overlay method.
We home audio folks who can stand latency have it easy....just use complementary linear pass xover between sub and main (and sufficient taps!)
Live sound folks who have to stick with IIR, are the ones who have to work,....much harder to get phase rotations to overlay than straight flat lines.
Hope this all made sense/ helps....