[ekovalsky]

The bottom graph shows the measured frequency response of the main channels, digitally tri-amped with crossovers & correction programmed into each of the three amps with TACS.  Measurements obtained in room at 4 m, at the listening position..

The upper graph shows the time domain performance.  Speakers are symmetrically positioned in a symmetrical room, and microphone is exactly centered between them.  Note the almost perfect channel tracking until the later echoes start arriving.

[ekovalsky]

Here you can see the combined frequency response of the main channels and subwoofers, and the 70hz, 36dB/octave high and low pass filters implemented in the RCS 2.2X.  

The green line is the target curve, as you can see it follows the measurements very closely other than some I added to taste -- basically a modest shelf in the bass tapering off around 300hz, enrichening the tonal balance; a slight rise 600-1800hz, giving improved presence to vocals; and a slight dip 1800-4000hz which nudges the soundstage back a bit.

Subwoofers levels are set to -3dB at 20hz.

At 53hz there is a big dip in the response of both the mains and subs. The bass recovers below this, hence it is a room null.  Its frequency seems to be dependent on height.  Fortunately it is in a frequency range where I can work around it with the crossovers.  Flanking this are room modes at about 40hz and 77hz, which are easily handled.

[ekovalsky]

Finally, this is the somewhat cryptic TacT dual domain.  On the bottom is the frequency response of the correction filters.  Basically this shows how much attenuation or potentiation of the signal is given at any particular frequency to convolve the measurement into the target.  Filters for both the subwoofers and main channels are shown.  Notice the significant attenuation at 40hz, corresponding to the major room mode.  And notice that elsewhere the curve is very close to flat -- meaning that the RCS isn't doing much work!  And that is a good thing as it preserves bits.  High levels of correction (I've used up to 25dB in a former pair of speakers) result in loss of resolution from bit loss.

Output levels are controlled by each amplifier.  Since the S2150 attenuates by lower voltage to the power rails, no bits are lost.  The midrange and tweeter amps are set -10dB relative to the bass amp, since correction requires headroom and the greatest levels of correction are obviously in the bass.  Thus I use the RCS as a digital preamp to control volume, rather than letting it pass the volume settings off to the amps.  My comfortable listening level is -6dB to 0dB, and with the design of the RCS and S2150 there is no bit loss at these levels.

Up top is the time domain analysis.  Like the TACS capture, this shows excellent time alignment and channel balance.  The impulse heads down before up, this is a function of the mic or mic preamp in the TacT.  All drivers are run in positive polarity with this setup, which uses LR4 crossovers at 350hz and 2.7khz.

[JoshK]

Do you know why the impulse is centered around 17ms?  Is it because of added delay in the system?  Looks like there if very little pre-ring according to that measurement.  That is good.

[ekovalsky]

Do you know why the impulse is centered around 17ms?  Is it because of added delay in the system?  Looks like there if very little pre-ring according to that measurement.  That is good.

The 17ms number is easy to explain.  Speakers are about 4m from microphone.  So it takes about about 11 ms to travel from speaker to mic.  The rest is explained by the latency of the TacT amps, because of their DSP filters.  The woofer, using Fs=3khz filters, is the slowest.  So extra delay is added to the midrange, using Fs=12khz filters, and particularly the tweeter, using Fs=96khz filters, to time align the system.

[JoshK]

thanks, that make sense.

[zybar]

Looking great Eric!

Good to see that all that hard work and effort is paying off with some great sound.

As per your recommendation I made the slight bump in the midrange and a slight roll-off in the highs.  THis improved clarity and timbre as well as moving the soundstage back a bit.  

Thanks!

George

[ekovalsky]

Do you know why the impulse is centered around 17ms?  Is it because of added delay in the system?  Looks like there if very little pre-ring according to that measurement.  That is good.

You have the choice of using zero phase or minimum phase crossovers.  Zero phase suffer from ringing, the minimum phase do not !  Sound is definitely better with the minimum phase filters, undoubtedly for this reason.

[ekovalsky]

Looking great Eric!

Good to see that all that hard work and effort is paying off with some great sound.

As per your recommendation I made the slight bump in the midrange and a slight roll-off in the highs.  THis improved clarity and timbre as well as moving the soundstage back a bit.  

Thanks!

George

Work and effort aren't that hard, but it has been time consuming.  Much like when I first got the TacT equipment a year ago.  But I feel I have mastered TACS in the past week, so I can develop a new project from scratch in about an hour now.  After spending all week learning it, of course a step-by-step guide just got translated and published -- it's available on Yahoo group TAUGSOA2.



A little judiciously applied EQ can definitely make things sound better!  I have a curve saved with a bigger dip (about 3dB) from 1250-5000hz that works wonders with overly forward recordings.

[ctviggen]

Why do the subs produce sound up to almost 500 Hz?  You're getting close to having voice in the subs (voice typically begins around 500 Hz:

http://www.transom.org/tools/editing_mixing/200402.voiceprocessing.html, although we used to use a high end of 4kHz).

Why is there such a dip at 20 Hz?  I would think that those massive subs would produce more output at 20Hz.  

I take it that the top graph of the dual domain shows the impulse response of the filters used for correction.  How does the TACT actually perform this operation?  Are the actual correction functions performed in the time domain or frequency domain?  In other words, are they convolving in the time domain or multiplying in the frequency domain (or vice versa)?  What type of filter is used for the correction filter(s)?  Can you draw a diagram (or give the equation) of the correction filter?  How many poles does it (they) have?  If the the TACT is using two correction filters (one for low and one for high frequencies), then each would have its own impulse response.  Is the graph of impulse response of the correction filter for one or both correction filters?  How do they take into account the response of one correction filter when designing the other?  How are these low and high frequency correction filters actually applied in terms of frequency?  You show overlapping frequency response for the two filters.  Therefore, there should be some splitting of the signal (i.e., crossover filters) in order to determine two signals (low f and high f) that are then applied to each of the correction filters.  Each of those signals would still overlap.  They somehow have to take into account both low and high correction filter response (and phase).  For instance, the f response of the two correction filters show that both add to the response at about 60 Hz.  Therefore, they must have some way of figuring out what the end result will be.  What is the mathematics of that, particularly as it relates to phase (especially when one considers that not only are the subs and mains physically separated but also are made of many individual drivers, each with its own distance to the mic; I assume they use a gross approximation for phase, since they can't really correct the phase of each driver but only of the subs or mains).  I assume that correcting phase has to be done in a two step process -- the phase response of the correction filter must be itself corrected and the phase response of the subs/mids would also have to be corrected.  When is this done and how?

[ekovalsky]

Why do the subs produce sound up to almost 500 Hz?  You're getting close to having voice in the subs (voice typically begins around 500 Hz

The graph shows the frequency response of the subs run with a low frequency pulse that extends to 400hz.  The crossover (shown in pink) has not yet been applied.  When the correction is calculated from the measurement,  the target curve is adjusted to incorporate the selected crossovers.  With the filters I use, response in the subs is down -6dB at about 64hz and falls off 36dB/octave beyond that.  So at 128hz they're down 42dB.  And the inverse goes for the mains below 64hz, where they naturally roll off since the woofers are sealed.  Trust me, you do not hear any voices coming from the subs.  They blend in so well with the mains that you do not hear them at all, until you put in some program material that has massive LF signal and then they make themselves known to everyone in the house

Why is there such a dip at 20 Hz?  I would think that those massive subs would produce more output at 20Hz.

Well, the subs are driven by a Crown K2 with adjustable input sensitivity.  I have it set to match output with the mains after correction in the amps, i.e. -3dB at 20hz.  I could add add another 12dB output to the subs which would raise their response (blue and purple traces) cruves up quite a bit, i.e. +9dB at 20hz, without changing the shape of the curve.  But why would I want to do that ?  The TacT would just have to attenuate it back down to match the output of the mains. Also there is a dominant room node at 40hz and another at 77hz which exaggerate the output at those frequencies, making the bottom end roll off look steeper than it actually is.  Like any woofer, output at 20hz is lower than at 400hz.  This doesn't matter, since output beyond the bottom two octaves is eliminated by the crossover.  Only thing that matters is potential output at the frequencies you want the subs to reproduce.

Compare the output of these subs to that of another "full range" speaker I used to own, measured at the same distance in the same room.  Of course I wouldn't expect them to have the output sensitivity and extension of these huge sub towers, but in their case the low bass falls "off a cliff" below the lower room node of 40hz, despite having two large carbon fiber woofers, one with oversized motor, and a large passive radiator.  And in case anyone was going to ask, adding or removing a few milligrams of putty (even a few grams of putty) made no difference in the measurements.

I take it that the top graph of the dual domain shows the impulse response of the filters used for correction.  How does the TACT actu ...

The top graph of the DD screen shows the impulse of the measurements, not the correction.  There are 10 buffers loaded with data, the upper graph shows 5 & 6 while the lower graph, which does show the correction filter in the frequency domain, shows 7-10.  I could load the correction filter in the time domain and post a screencap of that.

As to have the TacT convolves its filters... you'd have to ask Boz himself !  And I have a feeling he won't be willing to share that information.  One thing I know is that correction resolution is dependent on frequency, below 400hz Fs=3khz, 400-3khz Fs=12khz, above 3khz Fs=96khz.

[jermmd]

The green line is the target curve, as you can see it follows the measurements very closely other than some I added to taste -- basically a modest shelf in the bass tapering off around 300hz, enrichening the tonal balance; a slight rise 600-1800hz, giving improved presence to vocals; and a slight dip 1800-4000hz which nudges the soundstage back a bit.

Eric,

I am playing around with my Tact and would appreciate your opinion on the target curve I should be aiming for. How many decibels off flat are your adjustments? Shouldn't the target curve be independent of the speakers and their meassured response? I realize trial and error will give me results but surely your experience will help.

Thanks,

Joe M.

[ekovalsky]

Eric,

I am playing around with my Tact and would appreciate your opinion on the target curve I should be aiming for. How many decibels off flat are your adjustments? Shouldn't the target curve be independent of the speakers and their meassured response? I realize trial and error will give me results but surely your experience will help.

Thanks,

Joe M.

Joe,

I'd suggest a 2-3dB rise in the target below 125hz, tapering to flat at 315hz.  Then a 1-2dB rise from 600-1.8khz and a 1-2dB dip from 1.8khz to 4.5khz.  Above that, the target should follow the natural response of your speaker.

With the RCS I am currently getting +1 / -1.5 dB between measurement and target curve, but this is because most correction is being performed in the amplifiers.  Before I was tri-amping with TACS, I was had about +1 / -10dB between measurements and target on the RCS with all the large deviations in the bass.  With the former speakers I was about +1 / -20dB with major deviations from measured in bass and midrange.

Ideally your correction filter should not go beyond 0dB.  But it is okay to go +1 to +3dB at a few frequencies as long as this is in very narrow bands.

Feel free to send me screencaps or target curve files, I'd be happy to look at them and make suggestions.  zybar would be a good resource too since he is familiar with TacT and has the Salk speakers also.

[zybar]

Joe,

I think it was correction number 7 where I basically did what Eric suggested.

Copy that correction curve and just adjust it from 4.5k up to follow the actual measurement.  

Give me a call if you have any questions.

George

[jermmd]

Thanks guys. I'm working for the next couple of days but I'll try the suggestions out next week.

[ctviggen]

I have to apologize to Eric.  I think what he's learned about the TACT is impressive, as is his system.  My comments have been inappropriate at best and childish at worst.  I won't interfere with Eric's postings in the future.