A detailed 2.0 with 2 subs correction- walk through

Omid

Member
Thread Starter
Joined
May 28, 2017
Messages
131
I thought I would do a little write up on my present setup in case it helps anyone else with a similar situation. My method is convoluted, but it gives me the best step response I ‘ve ever eked out of my system. Also, if I’ve done anything stupid, perhaps someone can let me know !


I have two main speakers, and two subwoofers, for playing stereo music (so a 2.0 setup in Audiolense). I have set the crossover point between the speakers and the subwoofers at 69 Hz. The crossover is asymmetrical. The speakers don’t play much below 50 Hz so there’s a steeper roll off for the speakers (2 octaves). The subwoofers can play much higher than 69 Hz, so there’s a gentler roll off for the subwoofers (3 octaves). Rather than dedicating one sub for each speaker, I have each subwoofers playing both channels (ie in mono). They cover the same frequency range. This allows me to fill room nulls a little bit and decrease distortion.
53300
53301


Taget and XO diagram:
53302


Target overlayed on uncorrected frequency response.
53315


Here are the correction filters. I fiddle with the target to try to avoid having a major spike at one specific frequency (it results in a high gain penalty for the sake of a single frequency, especially if <20Hz or >20,000Hz).
53309
53310


The difficulty is that Audiolense does not allow you to sweep the same channel more than once (here I need each sub to have a sweep for the right signal, and again for the left signal). To get around that, I have a special routing in my digital analogue converter. Signals in channels 2 and 4 both go to the same subwoofer (the one on the right side of the room), and channel 3 and 5 both go to the left subwoofer. So, the signal for the left sweeps goes to channels 2&5 (R+ L Subs) for low frequencies and channel 0 (L main) for 69Hz and above. In the measurement window, one sub is labelled “bass”, the other sub “midrange”, the mains are called “tweeter”. Despite the funny naming system, the “bass” and “midrange” play the same range 10-69Hz, the “tweeter” 69-20,000Hz.
53303




I have not separated the tweeter from the woofer on my speakers for this setup. In the past, I have removed the bridge between tweeter and woofer at the back of the speaker and had Audiolense oversee the crossover. But since I did not remove the internal crossover, there was no advantage that I could hear or see in my measurements. I don’t feel like opening the speakers to bypass the crossover. My amps working as mono blocks have 6dB more gain than when playing two channels each (tweeter and woofer). Also, the tweeter seems to be much more sensitive than the woofer. I think this sensitivity mismatch will cause a big correction in AL, which means a big penalty in overall gain. Both facts would decrease the maximum loudness I can play at. Being able to play scarily loud is imperative : )

For reference, these are the measurement before Audiolense correction (left channel shown, 2 subs, 1 speaker):
53316
53317


For the measurements, I decided to try something new. I did sweeps of five seconds, repeating the sweep 10 times, but without moving the microphone at all. I then selected multi seat correction. I figured this would allow me to average 10 identical measurements, reducing any random errors and noise. Not sure if that’s a smart idea. It did give me better results than what I’ve had before (i.e. better simulated and measured step responses). You can see in the pictures, all measurements are within +/-0.5dB of each other, in the 20-20,000Hz range.
53304
53305



For the corrections, I chose to add a mid-band correction. That way all the low frequencies from 20 to 300 Hz are done with a higher resolution (8 cycle window, 6 cycle sub-window), and then after 300 Hz the resolution decreases gradually to 2 cycle window, 1.008 cycle sub-window. I need to click 'prevent bass boost' even though the target is pretty steep in low frequencies, if I want to avoid a big correction under 10Hz.
53306





These are the simulated frequency and step responses.
53307
53308

There is a bit of pre-ringing but really low. Selective pre-ringing prevention makes things worse.




These are the smoothed measured responses in REW (sorry the freq response rolls off in the high freq as I had a slightly different target with this set of measurements, but all the rest is the same).
53314
53313








I included a sim freq response and step response without multi seat, as you can see the results are not as good.
53311
53312




The system sounds good. The image is well centered. The measurements look pretty good to me.

I hope his helps someone. Let me know if I can do anything differently to improve my results.
 
Last edited:

Ofer

Member
Joined
Mar 15, 2021
Messages
244
More  
Preamp, Processor or Receiver
miniDSP 4x10hd
Main Amp
Emotiva XPR200 midrange amp
Additional Amp
Crest audio 2001A bass amp, Crest audio 8002 sub
Other Amp
Rotel RA930ax twitter amp
Computer Audio
Sony Bravia android TV
DAC
RME FF400
Universal / Blu-ray / CD Player
Marantz original 5E CD
Front Speakers
Andromeda MkII
Subwoofers
18" OEM powered subwoofer, 18" Martycube Dayton A.
Screen
Sony bravia 65XF9005
Streaming Subscriptions
Deezer HiFi
Other Equipment
HP i5 running W10, HLC convolver for Audio Lense filters
Hi Omid,

Thanks for this, you are very helpful to the community for a while now. A few questions:
1. You have the TTD correction per driver and TTD correction ticked. From what what @Mitchco wrote in his walk through this could only be achieved when the drivers are corrected separately by AL. Which if I understand correctly is not the case here. How did it work for you?
2. If I remember correctly. You once tried to use the summed input 6/7 in the audio card to treat the sub as a 2way speaker and specify both hpf and lpf. How did that work? Why (and if) did you give up on that idea?
3. What data caused the decision to choose a 300hz mid frequncy and not higher or lower?
4. What smoothing is used for the REW measurement?
Thanks for your contribution to the forum so far.
 
Last edited:

Omid

Member
Thread Starter
Joined
May 28, 2017
Messages
131
Hi Omid,

Thanks for this, you are very helpful to the community for a while now. A few questions:
1. You have the TTD correction per driver and TTD correction ticked. From what what @Mitchco wrote in his walk through this could only be achieved when the drivers are corrected separately by AL. Which if I understand correctly is not the case here. How did it work for you?
2. If I remember correctly. You once tried to use the summed input 6/7 in the audio card to treat the sub as a 2way speaker and specify both hpf and lpf. How did that work? Why (and if) did you give up on that idea?
3. What data caused the decision to choose a 300hz mid frequncy and not higher or lower?
3. What smoothing is used for the REW measurement?
Thanks for your contribution to the forum so far.

Thank you for the comments. I'm no expert, so my solutions are not necessarily the best, just what I've come up with so far...

1. Each of my virtual speakers are composed of 3 drivers: 2 subs and a main, with a crossover at 69 Hz. It is the similar to having a woofer and tweeter. In other words AL is doing a TTD per driver (each sub, and the main speaker).
2. Summing the 2 subs doesn't allow AL to individual correct each sub. It only correct the sum of the 2 (so a more blunt tool).
3. My Subs' output is way down by 300Hz, so I figured less detail was needed after that. I did try a range of frequencies though, and this is the one that gave the best curves. I think if I had a better understanding of the maths Audiolense uses, I would choose my variables more smartly. I feel like I'm semi-blindingly changing variables to get the best freq and step response. Sometimes a little change induces a huge unexpected change in results. puzzling...
4. I used psychoacoustic smoothing for the freq response, because otherwise I obsess over variations that I can't actually hear. Like the narrow band 2.2 dB difference at 87Hz, in the graph below. Here's the REW measured freq response with 1/6 octave smoothing:
53319
 

kevinzoe

Member
Joined
Jun 10, 2022
Messages
39
More  
Preamp, Processor or Receiver
PrimaLuna DiaLogue Premium preamp
Main Amp
PS Audio BHK Signature 250 stereo amp
Additional Amp
Valve Amplification Company 30/30 Renaissance
Computer Audio
JRiver
DAC
exaSound e48 (8 channels)
Front Speakers
Acoustic Zen Crescendos
Subwoofers
Pair of Rythmik F15's
Other Equipment
Cables by: TEO Audio, Cardas and AudioSensibility
Omid - Great write-up thanks!

Here's a few ideas to try on for size based on my limited experience with a very similar setup . . .

> CORRECTION AMOUNTS (see correction chart)

- your red sub volume seems too low relative to the blue sub volume at the mic position. I suspect this to be the case because there is more correction needed for the red sub and less for the blue sub. Try and get both sub's correction amounts to be the same and this should help bring down the large correction peak at 40Hz for the red sub (and other peaks). I may misunderstand the red and blue lines . . .

- to avoid the largish correction boost between 10-20hz for the two subs, I would drag your target curve marker for 10hz down much lower say -80d and rerun the filter generation to see if that doesn't reduce the correction amounts.

. the huge peak at about 18kHz can be reduced IF you tailor your target curve after 10kHz to match the natural tweeter roll off of your mains. Rerun the Generate correction until the natural roll off matches the target curve.

- I bet that with the 3 above changes, you will see that your target curve and predicted frequency response curves will shift upwards, meaning that you've reduced the "filter insertion loss" and in turn the amplitude of your impulse response will get larger which means that your music will sound more dynamic. (I'm assuming if you like your music loud, then you might also like it with a healthy dose of dynamics.)

- another way to reduce filter insertion loss and improve impulse response is to use a lower filter length in the Correction Procedure Designer from 131,072 down to 32,000 or even 16,000. What this change does is reduce the amount of correction applied so your predicted freq response isn't quite as smooth BUT you gain impulse response amplitude and music sounds more dynamic (which is important to me because if you consider music from a radio, a great stereo, or a live show, the biggest difference to me is dynamic contrasts - barely any with radio, moderate amounts with a stereo, and wild amounts possible with a live show.) I think you like to play your music loud so these suggestions may play into that . . .

> CORRECTION PROCEDURE DESIGNER

- The "Measurement & Correction Window" values for the "Cycles before & after peak" should be 2-3 amounts higher than their corresponding values in the "True Time Domain Sub Window." What's askew in your picture are the values of the 48,000Hz (FS / 2) -> raise the 2 up to 3 or 4 in the Measurement & Correction Window.

. try reducing the Max Correction Boost from 6dB down to 5 or 4 and see how the predicted Freq Response changes and how much more Impulse Response amplitude you gain. I found that this is a personal trade-off between as flat a freq response you can get and the killing of dynamics, so you may need to give up some of one to gain more of the other.

> LOW PASS FILTER SLOPE OF SUB

- I see a largish null at about 88Hz in the red line of your "simulated frequency and step responses" chart. (Too bad your charts weren't numbered so I could point you to them more easily.) You may be able to reduce this null amount by making the sub's slope steeper, so try going from a 3octave width to a 2.75 or 2.5 or 2.25 or a 2 and rerun the filter and generate correction again to see if it disappears.
 

Ofer

Member
Joined
Mar 15, 2021
Messages
244
More  
Preamp, Processor or Receiver
miniDSP 4x10hd
Main Amp
Emotiva XPR200 midrange amp
Additional Amp
Crest audio 2001A bass amp, Crest audio 8002 sub
Other Amp
Rotel RA930ax twitter amp
Computer Audio
Sony Bravia android TV
DAC
RME FF400
Universal / Blu-ray / CD Player
Marantz original 5E CD
Front Speakers
Andromeda MkII
Subwoofers
18" OEM powered subwoofer, 18" Martycube Dayton A.
Screen
Sony bravia 65XF9005
Streaming Subscriptions
Deezer HiFi
Other Equipment
HP i5 running W10, HLC convolver for Audio Lense filters
Hi Omid,
What convolver do you use?
 

Omid

Member
Thread Starter
Joined
May 28, 2017
Messages
131
Omid - Great write-up thanks!

Here's a few ideas to try on for size based on my limited experience with a very similar setup . . .

> CORRECTION AMOUNTS (see correction chart)

- your red sub volume seems too low relative to the blue sub volume at the mic position. I suspect this to be the case because there is more correction needed for the red sub and less for the blue sub. Try and get both sub's correction amounts to be the same and this should help bring down the large correction peak at 40Hz for the red sub (and other peaks). I may misunderstand the red and blue lines . . .

- to avoid the largish correction boost between 10-20hz for the two subs, I would drag your target curve marker for 10hz down much lower say -80d and rerun the filter generation to see if that doesn't reduce the correction amounts.

. the huge peak at about 18kHz can be reduced IF you tailor your target curve after 10kHz to match the natural tweeter roll off of your mains. Rerun the Generate correction until the natural roll off matches the target curve.

- I bet that with the 3 above changes, you will see that your target curve and predicted frequency response curves will shift upwards, meaning that you've reduced the "filter insertion loss" and in turn the amplitude of your impulse response will get larger which means that your music will sound more dynamic. (I'm assuming if you like your music loud, then you might also like it with a healthy dose of dynamics.)

- another way to reduce filter insertion loss and improve impulse response is to use a lower filter length in the Correction Procedure Designer from 131,072 down to 32,000 or even 16,000. What this change does is reduce the amount of correction applied so your predicted freq response isn't quite as smooth BUT you gain impulse response amplitude and music sounds more dynamic (which is important to me because if you consider music from a radio, a great stereo, or a live show, the biggest difference to me is dynamic contrasts - barely any with radio, moderate amounts with a stereo, and wild amounts possible with a live show.) I think you like to play your music loud so these suggestions may play into that . . .

> CORRECTION PROCEDURE DESIGNER

- The "Measurement & Correction Window" values for the "Cycles before & after peak" should be 2-3 amounts higher than their corresponding values in the "True Time Domain Sub Window." What's askew in your picture are the values of the 48,000Hz (FS / 2) -> raise the 2 up to 3 or 4 in the Measurement & Correction Window.

. try reducing the Max Correction Boost from 6dB down to 5 or 4 and see how the predicted Freq Response changes and how much more Impulse Response amplitude you gain. I found that this is a personal trade-off between as flat a freq response you can get and the killing of dynamics, so you may need to give up some of one to gain more of the other.

> LOW PASS FILTER SLOPE OF SUB

- I see a largish null at about 88Hz in the red line of your "simulated frequency and step responses" chart. (Too bad your charts weren't numbered so I could point you to them more easily.) You may be able to reduce this null amount by making the sub's slope steeper, so try going from a 3octave width to a 2.75 or 2.5 or 2.25 or a 2 and rerun the filter and generate correction again to see if it disappears.

Wow !! Thank you Kevin for the great feedback. Lots of stuff to try.

I think this type of discussion can helps us all.

My amps don't have super high gain (23dB), and my speaker are 91dB/2.83Vrms sensitive, so I do like minimizing losses. The REW measurement I posted used a -26dB dBFS signal and gave ~80dB at my seating position, so I have a max possible output around 105dB which is ok-ish, I think (22 dB dynamic range at 83dB average listening level).

So I'll start with your suggestions on modifying my target curve to avoid a big correction around 10 Hz. My pic above did not show the full range of my target curve. Here's a better picture (chart 1):
53324

as you can see I did go pretty low. But I did modify my target as you suggested (chart 2):
53325


Strangely, if I don't select 'prevent bass boost' I still get a big bump on the R channel around 10Hz (chart 3):
53326


Once I engage prevent bass boost, the 2 targets yield identical results. I don't mind using this option though as it does not seem to cause any problems.

With respect to your suggestions to decrease the correction bumps at low and high frequencies, they are good points. My perspective is that there is no getting rid of the correction on my main speakers in the 200-300 Hz. That's my rate limiting step. Since my correction at high and low frequency exceed the correction at 200-300 Hz by 1 dB or so, by working on the highs and lows I may be able to squeeze 1 dB more. Worth a try !

The idea of decreasing filter length to affect correction amounts is something that hadn't occurred to me (I onlyt thought of it as a means of increase low freq correction resolution, at the possible cost of increased computation time, and signal manipulation artefact). So I gave it a shot. In my case, as you predicted the correction amplitude in the low frequency diminished. The overall result did not pan out well though. Here are the result:
decrease in correction amplitude (chart 4):
53329

vs what it used to be (chart 5):
53330


but not a great result (chart 6 & 7):
53327
53328


Finally, I changed the window cycle length in high frequencies to 4 cycles instead of 2. The results did not change much (the high frequency have finer corrections, a less smooth curve (chart 8 &9)) :
53335

53336


2 cycle window for reference (chart 10 & 11):
53338

53337



I also tried reducing my Subs crossover from 3 octave to 2.5 and 2 octave. Unfortunately the 88Hz dip remained unchanged (same step response too).

I really appreciate all your feedback. I'll try to change my sub's volume a bit , and shave the high freq target, and play with the max correction boost, to see If I can get a bit less correction loss.

Cheers

Omid.
 

kevinzoe

Member
Joined
Jun 10, 2022
Messages
39
More  
Preamp, Processor or Receiver
PrimaLuna DiaLogue Premium preamp
Main Amp
PS Audio BHK Signature 250 stereo amp
Additional Amp
Valve Amplification Company 30/30 Renaissance
Computer Audio
JRiver
DAC
exaSound e48 (8 channels)
Front Speakers
Acoustic Zen Crescendos
Subwoofers
Pair of Rythmik F15's
Other Equipment
Cables by: TEO Audio, Cardas and AudioSensibility
. . . hmm interesting.

Can you try a few more things for me . . .?

(A) Can we fix the huge correction at the high end of the freq response by adding more target curve markers above 10kHz to map the target to the actual tweeter roll-off? You may need a few iterations to get it right after Generating Correction to show the target overlapping perfectly the predicted response. You might try ticking the "Prevent treble boost" box.

B) can you un-tick the "TTD correction per driver" found in the Correction Procedure Designer and re-run Generate Correction?

(C) May we see the filtered left and right channel curves please? I'd like to see what's going on at the crossover points and the sub's freq response.

(D) Chart 4 - what filter length and Max Correction amount in dB's were used? Those corresponding freq and IR charts 6 & 7 don't look favourable I'm afraid . . . The bass predicted response seem really bad for some reason. Did you go too low on the max correction and/or filter length settings?

(E) Regarding the sub's 88Hz null, if making the slope steeper (from 3 to 2) didn't work then keep trying 1.5 and 1 and if that doesn't work then try reducing the crossover frequency from 69 down to say 65 then 60 and re-run Generate Correction each time of course.

(F) did you experiment using differing window lengths of time to see if that might align the predicted IR with the target?

(G) as a last resort, is it possible to have one sub on the left channel and the other the right channel so they aren't actually mono? I'm wondering if this isn't throwing a wrinkle into things . . .
 
Top Bottom