Hi Hartono,
Thanks for your reply [post #1423]
The D-I filter (unlike digital interpolation filters often used in OS DACs) is phase linear, and all DAC chips still run in NOS mode. The D-I filter delays all frequencies by the same amount (group delay). The D-I filter group delay equals 22.6 uS.
The main reason I used multiple DAC chips is to increase both time and bit resolution.
If a NOS DAC has a resolution of 1 (16 bits, 44100 samples / second), the DI 8 has a theoretical resolution of 64 (19 bits, 352800 samples / second), and the DI 16 has a theoretical resolution of 256 (20 bits, 705600 samples / second).
By comparison a 16 bit OS DAC with 8 times oversampling only has a time resolution that's 8 times higher when compared to a NOS DAC.
So despite the fact D-I DACs are running in NOS mode, they provide much higher resolution (before filtering) than most 16 bit OS DACs.
Thanks for your reply [post #1423]
The D-I filter (unlike digital interpolation filters often used in OS DACs) is phase linear, and all DAC chips still run in NOS mode. The D-I filter delays all frequencies by the same amount (group delay). The D-I filter group delay equals 22.6 uS.
The main reason I used multiple DAC chips is to increase both time and bit resolution.
If a NOS DAC has a resolution of 1 (16 bits, 44100 samples / second), the DI 8 has a theoretical resolution of 64 (19 bits, 352800 samples / second), and the DI 16 has a theoretical resolution of 256 (20 bits, 705600 samples / second).
By comparison a 16 bit OS DAC with 8 times oversampling only has a time resolution that's 8 times higher when compared to a NOS DAC.
So despite the fact D-I DACs are running in NOS mode, they provide much higher resolution (before filtering) than most 16 bit OS DACs.
-ecdesigns- said:
Sorry, but:
No, no, no.
First, you have the linear interpolation, introducing errors that are magnitudes greater than the tiny LSBs of a 19/20bit DAC.
Second, the result of parallelling n DACs is a statistical problem.
Errors should cancel if n is a big number but in reality errors can go into the same direction and even add up.
And one single good chip can be much better.
Bernhard said:
In the accuphase dp-78 they put in // the dac and for this application it not seems to be a problem ?
They write:
The DP-78 uses six Analog Devices AD1955 delta sigma converters
in parallel, which affords an overall performance improvement by a
factor of 2.4 (�ã6)
Did I miss something ?
Hi Bernhard,
Thanks for your reply [post #1442]
Before you argue that linear interpolation errors are so bad, I will try to explain what's most important to obtain transparent sound quality.
If you attend a live performance, it doesn't matter if the instruments produce distortion, the direct sound and the slightly delayed reflections are 100% perfect and provide excellent detail about acoustic space. That's why it sounds so natural / transparent.
In order to re-produce this live sound accurately, the information about acoustic space is of eminent importance
Where is this information "hidden" in the audio signal? Just a simple test, add 2 sinewaves of nearly the same frequency, what do you observe? amplitude / waveform fluctuations. In other words, information about acoustic space is located in (minute) amplitude fluctuations. Also important is the time resolution of these fluctuations.
The higher the resolution of these fluctuations, the clearer the information about acoustic space gets, and the more transparent the perceived sound quality.
How can we effectively destroy this information about acoustic space?
Either by removing it all together (filtering) or distort / mask it by adding other frequencies.
- jitter, timing jitter translates to small amplitude variations in non-linear systems. These small amplitude variations will mask the information about acoustic space. Just test it, increase timing jitter and listen what happens.
- Filtering, first of all, filters tend to remove small sample-to sample amplitude fluctuations. The higher-order filters are used, the worst it gets. Now think again about brickwall filters. Next, filters are often non-linear, (FIR filters are phase linear and the non-linear interpolation algorithm doesn't count) this will translate timing jitter into amplitude fluctuations, again masking information about acoustic space.
Brickwall filters all have a certain amount of ripple voltage, again ideal for masking information about acoustic space.
- Non-linear components in the signal path, as they translate timing fluctuations into amplitude fluctuations.
Now back to the "crappy" D-I DACs with their "bad" linear interpolation and no analog corrective filter
The D-I filter will pass ALL small sample-to sample information that holds important information about acoustic space.
The D-I filter has a linear response, so it won't translate timing fluctuations into amplitude fluctuations.
The D-I filter increases resolution, reduces noise and enables the small amplitude fluctuations to surface.
The absence of a corrective analog filter, and the fully DC coupled design, result in a highly linear response, optimally preserving information about the acoustic space.
However, a lot of "damage" is already done by the necessity of using an anti-aliasing filter and a sampling clock (with jitter) during sampling. Bad mixing makes matters even much worse.
After extensive listening tests, both, good live recordings and Chesky audio recordings sound best, as they also hold most accurate information about acoustic space, and offer realistic dynamics.
Well according to Gausian distribution, the bit errors should concentrate around the ideal value (bell curve). This basically means that when multiple DAC chips are connected in parallel, there is a high probability that bit errors would be reduced.
TDA1541A chips have specified / verified bit errors, Philips TDA1541A datasheet page 7:
TDA1541A/N2 bit 1-16 EdL < 1 LSB
TDA1541A/N2/R1 bit 1-16 EdL < 2 LSB
TDA1541A/N2/S1 bit 1-7 EdL < 0.5 LSB
bit 8-15 EdL < 1 LSB
bit 16 EdL < 0.75 LSB
I am using the plain TDA1541A version, this has specified linearity errors of Edl < 1LSB
So the chips used, have a specified tolerance, this makes it even more likely that resulting bit errors would concentrate around the ideal value according to Gausian distribution.
Since I use a balanced design, I can correct errors even more effectively by swapping chips between the non-inverted and inverted DAC groups. The result can be verified by distortion measurements.
The TDA1543 has higher bit errors, so it's reasonable to assume that selecting chips with lowest distortion AND placing them in the right DAC groups is necessary to obtain low bit errors. This is basically no problem as TDA1543 chips aren't that expensive.
I already explained several times how the D-I system works. It's not simply paralleling DAC chips, like seen in many other designs. The multiple DAC chips used in the D-I DAC, provide increased (time) resolution, so the analog output filter could be removed, enabling linear response.
Thanks for your reply [post #1442]
Before you argue that linear interpolation errors are so bad, I will try to explain what's most important to obtain transparent sound quality.
If you attend a live performance, it doesn't matter if the instruments produce distortion, the direct sound and the slightly delayed reflections are 100% perfect and provide excellent detail about acoustic space. That's why it sounds so natural / transparent.
In order to re-produce this live sound accurately, the information about acoustic space is of eminent importance
Where is this information "hidden" in the audio signal? Just a simple test, add 2 sinewaves of nearly the same frequency, what do you observe? amplitude / waveform fluctuations. In other words, information about acoustic space is located in (minute) amplitude fluctuations. Also important is the time resolution of these fluctuations.
The higher the resolution of these fluctuations, the clearer the information about acoustic space gets, and the more transparent the perceived sound quality.
How can we effectively destroy this information about acoustic space?
Either by removing it all together (filtering) or distort / mask it by adding other frequencies.
- jitter, timing jitter translates to small amplitude variations in non-linear systems. These small amplitude variations will mask the information about acoustic space. Just test it, increase timing jitter and listen what happens.
- Filtering, first of all, filters tend to remove small sample-to sample amplitude fluctuations. The higher-order filters are used, the worst it gets. Now think again about brickwall filters. Next, filters are often non-linear, (FIR filters are phase linear and the non-linear interpolation algorithm doesn't count) this will translate timing jitter into amplitude fluctuations, again masking information about acoustic space.
Brickwall filters all have a certain amount of ripple voltage, again ideal for masking information about acoustic space.
- Non-linear components in the signal path, as they translate timing fluctuations into amplitude fluctuations.
Now back to the "crappy" D-I DACs with their "bad" linear interpolation and no analog corrective filter
The D-I filter will pass ALL small sample-to sample information that holds important information about acoustic space. The D-I filter has a linear response, so it won't translate timing fluctuations into amplitude fluctuations. The D-I filter increases resolution, reduces noise and enables the small amplitude fluctuations to surface. The absence of a corrective analog filter, and the fully DC coupled design, result in a highly linear response, optimally preserving information about the acoustic space.However, a lot of "damage" is already done by the necessity of using an anti-aliasing filter and a sampling clock (with jitter) during sampling. Bad mixing makes matters even much worse.
After extensive listening tests, both, good live recordings and Chesky audio recordings sound best, as they also hold most accurate information about acoustic space, and offer realistic dynamics.
Well according to Gausian distribution, the bit errors should concentrate around the ideal value (bell curve). This basically means that when multiple DAC chips are connected in parallel, there is a high probability that bit errors would be reduced.
TDA1541A chips have specified / verified bit errors, Philips TDA1541A datasheet page 7:
TDA1541A/N2 bit 1-16 EdL < 1 LSB
TDA1541A/N2/R1 bit 1-16 EdL < 2 LSB
TDA1541A/N2/S1 bit 1-7 EdL < 0.5 LSB
bit 8-15 EdL < 1 LSB
bit 16 EdL < 0.75 LSB
I am using the plain TDA1541A version, this has specified linearity errors of Edl < 1LSB
So the chips used, have a specified tolerance, this makes it even more likely that resulting bit errors would concentrate around the ideal value according to Gausian distribution.
Since I use a balanced design, I can correct errors even more effectively by swapping chips between the non-inverted and inverted DAC groups. The result can be verified by distortion measurements.
The TDA1543 has higher bit errors, so it's reasonable to assume that selecting chips with lowest distortion AND placing them in the right DAC groups is necessary to obtain low bit errors. This is basically no problem as TDA1543 chips aren't that expensive.
I already explained several times how the D-I system works. It's not simply paralleling DAC chips, like seen in many other designs. The multiple DAC chips used in the D-I DAC, provide increased (time) resolution, so the analog output filter could be removed, enabling linear response.
-ecdesigns- said:
That only means that the transfer curve has a max. deviation of 1 LSB from ideal at zero crossing when only looking at the DLE.
Let me give you an example:
I first select PCM56s for being perfect after MSB adjust.
That means the selected chips are very linear exept DLE error.
( Bad chips can not be corrected )
In that selected batch there are three kind of chips.
1) requires the pot to be turned somewhere to the left.
2) requires the pot to be turned somewhere to the right.
3) everything is fine exactly in the middle, you may forget about the pot.
Lets say the batch contains 32 chips.
From that we built a 16 parallel chip / channel DAC without LSB adjust.
1 chip is type 3
21 chips are type 1
10 chips are type 2
What will you do ?
USB/DI2S
Dear John (ECdesigns)
Thank you for giving me a tour last Sunday at your place.
I am still impressed by your your lab and DI8dac --> amp --> resonators.
I received this morning the package with the USB/DI2S stuff.
Everything was well packed and complete and 2 hours after receiving it , the pc has seen a "audio DAC" !!.
The DI2S input is symetric.
Is it possible to use it as a asymetric input by only using
BCK+,WS+ and DATA+ and leave the negative connections NC !
Regards
Onno
Dear John (ECdesigns)
Thank you for giving me a tour last Sunday at your place.
I am still impressed by your your lab and DI8dac --> amp --> resonators.
I received this morning the package with the USB/DI2S stuff.
Everything was well packed and complete and 2 hours after receiving it , the pc has seen a "audio DAC" !!.
The DI2S input is symetric.
Is it possible to use it as a asymetric input by only using
BCK+,WS+ and DATA+ and leave the negative connections NC !
Regards
Onno
Bernhard said:
This is like crossing a street with eyes closed because there are not so many cars and there is a high probability not being hit by a car because there are not so many.
Wadia made in the late eighties a dac with paralleled PCM56's, and used their "bio algorythm digital filter"
Hi
ECdesigns, thanks for the reply...
It seems that measurements simply could been avoided...
For now i am implemented the Rv-ref resistor of 1350 ohm per TDA1543 unit (about +2.5Vdc at Vref pin) and about 2200 ohm for Riv resistor (+4.5Vdc at the output audio pins)... The sound is excelent, but i did not measured distorsion...
ECdesigns, thanks for the reply...
It seems that measurements simply could been avoided...
For now i am implemented the Rv-ref resistor of 1350 ohm per TDA1543 unit (about +2.5Vdc at Vref pin) and about 2200 ohm for Riv resistor (+4.5Vdc at the output audio pins)... The sound is excelent, but i did not measured distorsion...
Hi Ecdesigns,
Thanks for your reply (post#1440) and sorry for the late reply from my part. Too much work...
I see a lot of action since my last visit, but mainly vocal action
Why don't you guys try this and see for your selves, as this is an empirical hobby.
First of all, I think ecdesigns is not that lost when he analyses the main factors that influences sound reproduction/perception. Those points could be behind the fact that simpler NOS DAC's are musically satisfying even while they should not be such.
My mistake, I should have added that I use the optional high quality Tent Labs clock on that USB interface.
It does not sound jittery to me
Of course I will try your USB/DI2S receiver, when funds allow and if I get through the 48Fs problem...plus a little help to build a DI2S driver
I will have to study some papers, then...
Note that with the gyrator on digital PS the sound is not at all thin, au contraire but it does sound a little less extended than before, wich can be explained by the regs issue or lack of burn-in.
Thanks for that. I will have to find a power darlington with higher HFe or build my own, wich could be tricky as I read (potentially unstable). I read that Vc-e drop should be 1.4V; mine is higher; probably my 50VA transformer can't cope with the load? (Darlington heatsink gets hot)
Cheers,
M
Thanks for your reply (post#1440) and sorry for the late reply from my part. Too much work...
I see a lot of action since my last visit, but mainly vocal action
Why don't you guys try this and see for your selves, as this is an empirical hobby.
First of all, I think ecdesigns is not that lost when he analyses the main factors that influences sound reproduction/perception. Those points could be behind the fact that simpler NOS DAC's are musically satisfying even while they should not be such.
My mistake, I should have added that I use the optional high quality Tent Labs clock on that USB interface.
It does not sound jittery to me
Of course I will try your USB/DI2S receiver, when funds allow and if I get through the 48Fs problem...plus a little help to build a DI2S driver
I will have to study some papers, then...
Note that with the gyrator on digital PS the sound is not at all thin, au contraire but it does sound a little less extended than before, wich can be explained by the regs issue or lack of burn-in.
Thanks for that. I will have to find a power darlington with higher HFe or build my own, wich could be tricky as I read (potentially unstable). I read that Vc-e drop should be 1.4V; mine is higher; probably my 50VA transformer can't cope with the load? (Darlington heatsink gets hot)
Cheers,
M
Hi maxlorenz,
Thanks for your reply [post #1453]
OK, how is the Tent Labs clock used? is it used for reclocking BCK from the USB module?
I used a 2 x 9V 15VA torroidial transformer, rectified 9V ac, using 11DQ10 Skottky diodes and added 2 x 1000uF capacitor in parallel for smoothing. These capacitors can provide peak currents when necessary.
The gyrator might have higher output impedance than 2 x 1000uF capacitors in parallel, it probably won't be able to provide peak currents, this in turn could limit dynamics. This could explain why the DI 16 sounds a little less extended when using the gyrator.
In my opinion a gyrator isn't needed for the DI 16, as it has a fully balanced design / high CMMR.
I have a theory about "burn-in". This phrase is probably related to burning-in of tubes (getting optimal emission from the cathode coating), in relation to semiconductors and passive components it seems to make little sense.
At first I thought not much of it, but every time I soldered something (speaker crossover filters, I/V diff amps, resistive volume control) I noticed it didn't sound optimal at first, like a slight degradation in performance. But after a few days, the sound quality clearly improved, returned to normal. Note that the same components sounded OK before the soldering, so a component "burn-in" wasn't causing this.
So I suspect that "burn-in" is rather a settling effect of a fresh made solder joint. The strange thing is that even when a fresh solder joint has reached ambient temperature, performance still isn't optimal, after a few days it is.
So it might be a good idea to wait a few days after an assembly or mod, before serious listening sessions / comparisons.
Thanks for your reply [post #1453]
My mistake, I should have added that I use the optional high quality Tent Labs clock on that USB interface.
It does not sound jittery to me
OK, how is the Tent Labs clock used? is it used for reclocking BCK from the USB module?
I used a 2 x 9V 15VA torroidial transformer, rectified 9V ac, using 11DQ10 Skottky diodes and added 2 x 1000uF capacitor in parallel for smoothing. These capacitors can provide peak currents when necessary.
The gyrator might have higher output impedance than 2 x 1000uF capacitors in parallel, it probably won't be able to provide peak currents, this in turn could limit dynamics. This could explain why the DI 16 sounds a little less extended when using the gyrator.
In my opinion a gyrator isn't needed for the DI 16, as it has a fully balanced design / high CMMR.
I have a theory about "burn-in". This phrase is probably related to burning-in of tubes (getting optimal emission from the cathode coating), in relation to semiconductors and passive components it seems to make little sense.
At first I thought not much of it, but every time I soldered something (speaker crossover filters, I/V diff amps, resistive volume control) I noticed it didn't sound optimal at first, like a slight degradation in performance. But after a few days, the sound quality clearly improved, returned to normal. Note that the same components sounded OK before the soldering, so a component "burn-in" wasn't causing this.
So I suspect that "burn-in" is rather a settling effect of a fresh made solder joint. The strange thing is that even when a fresh solder joint has reached ambient temperature, performance still isn't optimal, after a few days it is.
So it might be a good idea to wait a few days after an assembly or mod, before serious listening sessions / comparisons.
Hi EC,
I'll pass on this one...
It is just an experiment and it served to demonstrate how implementation can modifie the performance of a well ingeneered project
I have two choices:
1) Return to my previous PS and bypass gyrator...
2) Put a bigger transformer with higer secondaries VAC
BTW, my PS has 50VA R-core TX with 9v second. and common type diode rectifiers each with // 47nF+1R followed by 3 *3300uF/16V Rubycon ZL.
I totally agree about the solder issue. People laugh at me when I comment that my 2%Ag solder takes at least 3 days to "sound" good. But other elements also can benefit from burn-in: signal level silver wire takes loooong to burn-in; OCC copper wire takes a little less to sound OK; BG caps can take eons, depending on capacity...etc.
Just IME
Thanks,
M
I'll pass on this one...
It is just an experiment and it served to demonstrate how implementation can modifie the performance of a well ingeneered project
I have two choices:
1) Return to my previous PS and bypass gyrator...
2) Put a bigger transformer with higer secondaries VAC
BTW, my PS has 50VA R-core TX with 9v second. and common type diode rectifiers each with // 47nF+1R followed by 3 *3300uF/16V Rubycon ZL.
I totally agree about the solder issue. People laugh at me when I comment that my 2%Ag solder takes at least 3 days to "sound" good. But other elements also can benefit from burn-in: signal level silver wire takes loooong to burn-in; OCC copper wire takes a little less to sound OK; BG caps can take eons, depending on capacity...etc.
Just IME
Thanks,
M
Bernhard said:
Will check it over Bernard, have an old test in a magazine. Have to search it in my old archives on the attic.
Wadia didn't want to show they used PCM56's because the printing was grinded off. The testers (Audio & Techniek) recognised the chip and checked it, only a PCM56 could be used then, 16 pins DIL
tubee said:
Found it Bernard: dutch mag. A&T Mai 1991 issue #17,
Wadia WT-3200 CD mechanism tested (with CDM1 ofcoarse) as decoder a common SAA7210.
Dac: Wadia X-32 Digimaster, has AT&T DSP chips and four pcm56's parallel. (they could see PCM5* on a chip, so thats why they knew.
http://www.audiocostruzioni.com/r_s/sorgenti/convertitori-antonio/Wadia x32-22.jpg
The later X-64-4 is also spoken of, not tested, it just came out then in 1991
Hi everybody
Here is a preliminary report of the DI16 DAC. This has been a nightmare week at my job(s) (yes, i do work for a living ) and there was no much time for listening to music.
Beforehand, I must state that these listening sesions were far from ideal:
1) I'm not a PC expert and "transport" is my big PC, windows XP, using Realplayer (I plan to use Ubuntu as my main system, when time allows). I will have to buy a Notebook...
2) My beloved one wouldn't let me put the PC on the listening room (5*6.7*2.7m)... who would blame her...so the PC is located on the next room, signal driven through 5m interconnects! (common type balanced, shielded, mic cable).
3) USB cable is a generic type, shielded cable, nothing fancy...I'm sure wire must be important here.
The DAC is the DI16 core PCB with components based on Ecdesigns BOM, safe: main 100uF/50V electrolytics are Rubycon ZL ( the ones that bypass the DAC chips (don't remember part Nº) are 220uF); +15v reg is 7815; R are 0.1% where encouraged.
I will have to take pics I think.
I solved the regs heatsink problem in a primitive way... you'll see.
USB receiver is Doede's (first version).
System: pream/active crossover is a Twistedpearaudio's Kookaburra-->Ampslab crossover at 1K2Hz (I recently added balanced output operation with THAT 1646 balanced drivers --> 4 modded UCD180 (LM4562) AC coupled I'm affraid--> activelly biamped Beyma 15KX on Autograph enclosures (I will soon varnish them to show some picks )
First of all, DI16 present the NOS' qualities of a non-digital sound, a lack of edginess and a relaxed and elegant charachter that allows me to forget about critizising and immerse on the musical experience.
The second thing that strikes is a sense of structure and "completeness" (I'm sure there is a good word for this in german) to the music, and a feelling of rightness of the tempos. In fact the music feels a little slower but this feelling is accompanied by a deeper insight: it is like listening to your favourite symphony with your favourite conductor; suddenly all become clearer and meaningfull. It's hard to explain
The soundstage is wider than before. If you now my type of speakers you will know that they have big spacial and dinamic scales. Well, DI16 produces a wider picture. Depth has not increased yet. Big dinamic fortissimi are fast, massive but have that softness of real instruments that never hurts: it is like a wind...
As John declared, there is a sense of increased detail retrieval, but not an "articial" type of detail, but one that is organic and natural. For example: violin high armonics, wood resonances and bow sounds are apparent and welcome; kettledrums and (how do you call the bigger drums?) produce that "Townnn" sound that you feel crosses the stage (maybe the spacial cues that Ecdesigns commented) and makes believe your at the live performance...
Highs are sweet and clean. Mids are colorful but not warm. Bass are extended and rich in armonics (you must have full-range speakers to feel it, I guess).
Yet, maybe due to the non-ideal situation mentioned above, I feel much more can be obtained from this project. A warmer midrange would be great: class A for opamps to try. I feel that relaxed sound is not entirelly natural: big jump to 16*4 towers next
That's for now.
Cheers,
M
Here is a preliminary report of the DI16 DAC. This has been a nightmare week at my job(s) (yes, i do work for a living
) and there was no much time for listening to music.Beforehand, I must state that these listening sesions were far from ideal:
1) I'm not a PC expert and "transport" is my big PC, windows XP, using Realplayer (I plan to use Ubuntu as my main system, when time allows). I will have to buy a Notebook...
2) My beloved one wouldn't let me put the PC on the listening room (5*6.7*2.7m)... who would blame her...so the PC is located on the next room, signal driven through 5m interconnects! (common type balanced, shielded, mic cable).
3) USB cable is a generic type, shielded cable, nothing fancy...I'm sure wire must be important here.
The DAC is the DI16 core PCB with components based on Ecdesigns BOM, safe: main 100uF/50V electrolytics are Rubycon ZL ( the ones that bypass the DAC chips (don't remember part Nº) are 220uF); +15v reg is 7815; R are 0.1% where encouraged.
I will have to take pics I think.
I solved the regs heatsink problem in a primitive way... you'll see.
USB receiver is Doede's (first version).
System: pream/active crossover is a Twistedpearaudio's Kookaburra-->Ampslab crossover at 1K2Hz (I recently added balanced output operation with THAT 1646 balanced drivers --> 4 modded UCD180 (LM4562) AC coupled I'm affraid--> activelly biamped Beyma 15KX on Autograph enclosures (I will soon varnish them to show some picks
)First of all, DI16 present the NOS' qualities of a non-digital sound, a lack of edginess and a relaxed and elegant charachter that allows me to forget about critizising and immerse on the musical experience.
The second thing that strikes is a sense of structure and "completeness" (I'm sure there is a good word for this in german) to the music, and a feelling of rightness of the tempos. In fact the music feels a little slower but this feelling is accompanied by a deeper insight: it is like listening to your favourite symphony with your favourite conductor; suddenly all become clearer and meaningfull. It's hard to explain
The soundstage is wider than before. If you now my type of speakers you will know that they have big spacial and dinamic scales. Well, DI16 produces a wider picture. Depth has not increased yet. Big dinamic fortissimi are fast, massive but have that softness of real instruments that never hurts: it is like a wind...
As John declared, there is a sense of increased detail retrieval, but not an "articial" type of detail, but one that is organic and natural. For example: violin high armonics, wood resonances and bow sounds are apparent and welcome; kettledrums and (how do you call the bigger drums?) produce that "Townnn" sound that you feel crosses the stage (maybe the spacial cues that Ecdesigns commented) and makes believe your at the live performance...
Highs are sweet and clean. Mids are colorful but not warm. Bass are extended and rich in armonics (you must have full-range speakers to feel it, I guess).
Yet, maybe due to the non-ideal situation mentioned above, I feel much more can be obtained from this project. A warmer midrange would be great: class A for opamps to try. I feel that relaxed sound is not entirelly natural: big jump to 16*4 towers next
That's for now.
Cheers,
M