TPA3116D2 Amp

A blind test would have been revealing, especially as I know what I think I heard in my test of the mod. Maybe xrk could rename the sound files without revealing the mod included or not - obviously in a different order. It might eradicate bias from both sides.

IMO, special files for blind test is unnecessary. If you want to know whether someone commenting on the sound really can hear differences, just ask an
ABX result with Foobar.

I downloaded the vocal (A) only and did a brief listening. First, I listen by paying attention to my "emotion/feeling" whether I can hear enjoyment, fatigue or distortion. Then I listen by paying attention into details.

The TDA is definately inferior. It has higher distortion, especially in the form of terrible sibilance.

Now, zobel versus no-zobel. Upon first listening I got a feeling that I like to listen to file without zobel. Then I listened by paying attention to details. They are different but hard to say which one is better. So I try to listen to the bass notes.

At around 0:35 there are 2 consecutive main bass strums. The second strum of the bass string in one file seemed "misstuned" to the music. It seems that the sound with zobel is more correct.

Overall, the clips are a LOT better than X's previous ones.
 
What is a surprise is that you didn't have the TI bootstrap snubber on dug0's. That is all.
Forget differential parts in 3116 outputfilter. Change STM outputfilter snubber to TI like if you want to improve the STM amp, as the designers of some of the STM chipsets mention, gnd for snubbers close together and as close as posible to amplifier gnd.
 
IMO, special files for blind test is unnecessary. If you want to know whether someone commenting on the sound really can hear differences, just ask an
ABX result with Foobar.

I downloaded the vocal (A) only and did a brief listening. First, I listen by paying attention to my "emotion/feeling" whether I can hear enjoyment, fatigue or distortion. Then I listen by paying attention into details.

The TDA is definately inferior. It has higher distortion, especially in the form of terrible sibilance.

Now, zobel versus no-zobel. Upon first listening I got a feeling that I like to listen to file without zobel. Then I listened by paying attention to details. They are different but hard to say which one is better. So I try to listen to the bass notes.

At around 0:35 there are 2 consecutive main bass strums. The second strum of the bass string in one file seemed "misstuned" to the music. It seems that the sound with zobel is more correct.

Overall, the clips are a LOT better than X's previous ones.

Interesting that you detect sibillance - because normally that ties into peaks in the freq response, but they are virtually identical, except for maybe 0.3dB less on some of the peaks in the midrange.

The sound clips sound better than the last round because they were all produced by the 10F/8424 in a specially designed and tuned FAST reference monitor. That is how good a reference monitor should sound - clean, smooth, flat response, no ringing, no nasty dips or peaks, with +/-2.5ms group delay from 35Hz on up. I was listening to it myself and also agree the SQ is superior to the setup used to do the Blind Comparison threads. I think my next blind comparison thread will have the 10F/8424, TG9FD, and TC9FD all in this cabinet. As they have similar frequency response curves it will come down to dynamics and CSD.
 
Last edited:
What is a surprise is that you didn't have the TI bootstrap snubber on dug0's. That is all.
Forget differential parts in 3116 outputfilter. Change STM outputfilter snubber to TI like if you want to improve the STM amp, as the designers of some of the STM chipsets mention, gnd for snubbers close together and as close as posible to amplifier gnd.

Yes, this amp never got the bootstrap snubbers or OSCON's because it was always in production use a main test amp. Just never got around to it. Now is a good opportunity I guess and I might even have before and after sound clips.

Soldering SMT's is always kind of one of those things where I have to work up the courage for it. Get the goggles, tweezers, small solder tip, etc. Now that it is all laid out, a bit easier to do.
 
At around 0:35 there are 2 consecutive main bass strums. The second strum of the bass string in one file seemed "misstuned" to the music. It seems that the sound with zobel is more correct.


Hi Jay,
Thank you very much for your feedback.
As you (I suspect), I am very excited on the next steps of this mod.
I am not sure I eard every details of the clips.
Can you please tell wich clip ? A ? B ? C?

Regards,

Christian
 
Wife gone out, annoying cow, and his in bed, so got a quoted period to listen some more.

What I "think" I've heard is that in all samples the "no zobel" tracks come bottom.

When it comes to the lower notes, the "3116 with zobel" wins over the 7492. As an example, with the samples A+B I can clearly hear more clarity and reverberation of the bass strings.

When it comes to the highs, I hear the "7492" win over the "3116 with zobel" samples. An example hear is the cymbals in track 2. Once again more clarity and detail.

Overall listening experience I think the "3116 with zobel" beats the 7492 as it loses out on the highs far less than the 7492 loses on the bass.

I wonder if the "3116 with zobel" could equal or better the "7492" with the addition of the bootstrap!!! (I mean in highs).

Might give listen again now.
 
Last edited:
Couldnt get the circuit for the 7492 right in spice, well, by the look of the voltage and phase lines I presume I didnt as they were crazy!

I decided to reduce the circuit to just 1 input and output channel to make life easier. Did 4 comparisons of basic 3116 / 3116 with bootstrap mod / 3116 with Zobel mod / 3116 with both bootstrap & Zobel mod.

Although all 4 comparisons look pretty similar, when zooming in, and doing the fft tests there seems significant differences in all 4.

Starting with the db & degrees of the parrallel voltage lines before seperating....
Basic 3116 circuit had 0db at 0degrees.
Bootstrap mod had -1db at 0degrees.
Zobel mod had -1.2db at 0degrees.

The Voltage and phase lines then seperate at a particular frequency.

Basic 3116 circuit Voltage at 166Hz / Phase 79Hz. (both align with drops in xrk's first & second FR graph. The 166Hz roughly collates with XRK's Harmonic distortion graph, under which he mentions an increase in HD.)

Bootstrap mod Voltage at 146Hz / Phase at 20Hz. (Not tested)

Zobel mod Voltage at 346Hz / Phase at 10Hz. (346Hz aligns with peak on 2nd FR graph. Interestingly, 346Hz is roughly the point where XRK's Harmonic Distortion graph shows it levelling from its fall, and the point it starts to show increasing peaks to the right.)

Both mods together Voltage at 487Hz / Phase at 74Hz
(Not tested)

With the fft graph I looked at both voltage lines and the time the impulse took to settle together again.
Basic 3116 circuit = 100us
Bootstrap mod = 120us
Zobel mod = 90us
Bootstrap & Zobel = 210us

(the result of Zobel mod being slightly quicker at the Impulse settling i.e response, seems to agree with XRK's finding on his test of speaker for real).

It was the very minute differences in voltage & phase lines I saw in my original graphs which made me look again at this.

From looking tonight, it looks like the phase seperating lower in frequency allows them to do so much more gradually. And the voltage lines seperating much higher up in frequency allows a sharper seperation of the two.
The Zobel has the widest freq range of those physically tested by xrk, between phase and voltage seperating from alignment with each other.
The values I noted above show that both bootstrap and zobel together increase that even further.
I have no idea if this difference makes a difference, but there are points at which it collates with XRK's test result graphs.
Another example of this is that with both the Basic 3116 spice graph, and the 3116 with Zobel mod, it is at 1kHz that both the Voltage and Phase lines start to increase their downward slopes. On XRK's HD graphs this frequency collates with the peaking in THD reducing to virtually nil until around 3500Hz.

BUT.....

it may look like the "potential sound differences" may be because the Zobel mod shows the quickest impulse settling time from the values. Which both mods together look to increase significantly.

It will be interesting to see if the "both mods" test in reality gives a worse sound. If the fact that the Voltage & Phase lines of spice seperating further apart collates with improved THD then it would improve sound higher up in frequency than 1000Hz - 3500Hz. At 3500Hz is where the Voltage and Phase lines cross on the spice graphs.
 
Last edited:
Jay,
I've found the part.
With my PC only I don't find a major difference except volume.

Don't worry, camelator. I have had long experience with blind test and always found I can hear what others can't. Honestly I didn't expect anyone else can hear the difference between zobel versus non-zobel (but it seems somebody else can). The TDA versus TPA anyhow is so obviously different that some people can hear it.

You know, it is very easy to perceive differences even when we listen to exactly the same copies. A Foobar ABX test result can usually be required in a controlled blind test, to prove that what we hear is real.
 
I finally got some time to listen to the sound clips myself. Interesting how the experience with headphones is very different than live with my own ears. I definitely could hear a difference with the 7492 - it just sounds brighter in the HF's even though the frequency response measures almost the same. Like Jay said, it is very obvious, the difference in sound between the 3116 and 7492. I think it is almost like the difference between a paper cone driver vs an aluminum cone driver. The Zobel vs non-Zobel mod, on the other hand, was very difficult to tell the difference with headphones for me. However, in person, I have a preference and it was easy to tell, but won't say that is until I post the blind comparison sound clips. I can't figure out why the live listening test I had an immediate preference but could not tell with the sound clips and headphones.
 
Last edited:
I can't figure out why the live listening test I had an immediate preference but could not tell with the sound clips and headphones.

In order to detect difference, we need to see/hear problems. So non-ideal condition is required, to exaggerate the difference between good and not so good.

If you listen MP3 from 1m in front of the speaker, it sounds sufficiently good that you don't know if it has problems. But if you listen from outside, MP3 will have discontinuity when compared to better resolution files.

Nearfield listening such as headphones can be considered an ideal condition.
 
In order to detect difference, we need to see/hear problems. So non-ideal condition is required, to exaggerate the difference between good and not so good.

If you listen MP3 from 1m in front of the speaker, it sounds sufficiently good that you don't know if it has problems. But if you listen from outside, MP3 will have discontinuity when compared to better resolution files.

Nearfield listening such as headphones can be considered an ideal condition.

I have to agree with the near field comment. When I originally tested with one channel with zobel and other not, the detail was only picked out directly in front of zobel about 12inches at low volume. It was subtle, and did not stand out from further away between the non zobel channel.
I do think however that a dual channel zobel mod would show a slight improvement in soundstage separation due to the detail, and the harmonic difference your graph shows.
 
You checked for "soundstage" (earlier in this thread) with only one channel modified - how does this work?

Do you use a calibrated mic/dac/adc for the measurements (Havent found any hints so far), with always the same distance/angle?

It would be interesting to see n*measurements of the same (untouched) system over the day in one plot. (Like doing one measurement every two hours)

Or are you doing the measurements with a "class-d-reconstruction-filterbox" like this:

http://www.ti.com/lit/an/sloa107/sloa107.pdf

Just asking those questions, as i measure speakers and amps with "Holm impulse", and even slight variation of distance/angle/amplitude results in "errors" or variations.
 
You checked for "soundstage" (earlier in this thread) with only one channel modified - how does this work?

Do you use a calibrated mic/dac/adc for the measurements (Havent found any hints so far), with always the same distance/angle?

It would be interesting to see n*measurements of the same (untouched) system over the day in one plot. (Like doing one measurement every two hours)

Or are you doing the measurements with a "class-d-reconstruction-filterbox" like this:

http://www.ti.com/lit/an/sloa107/sloa107.pdf

Just asking those questions, as i measure speakers and amps with "Holm impulse", and even slight variation of distance/angle/amplitude results in "errors" or variations.

more subjective listening than measurement for me.
 
You checked for "soundstage" (earlier in this thread) with only one channel modified - how does this work?

Do you use a calibrated mic/dac/adc for the measurements (Havent found any hints so far), with always the same distance/angle?

It would be interesting to see n*measurements of the same (untouched) system over the day in one plot. (Like doing one measurement every two hours)

Or are you doing the measurements with a "class-d-reconstruction-filterbox" like this:

http://www.ti.com/lit/an/sloa107/sloa107.pdf

Just asking those questions, as i measure speakers and amps with "Holm impulse", and even slight variation of distance/angle/amplitude results in "errors" or variations.

I never said anything about sound stage, did I?
The measurements, as mentioned, were made with the speaker and mic untouched (position not moved or touched at all) - only amp was disconnected to perform the Zobel mod with a soldering iron or to swap amps to 7492. Yes, calibrated Dayton UMM-6 mic was used along with REW software.
 
Change STM outputfilter snubber to TI like if you want to improve the STM amp,

What "snubber"? The "zobel" as in zobel versus non-zobel of the TPA??

Such modification wont change much. The difference between TPA and TDA is too big, wont be affected by output zobel.

The zobel itself, imo doesn't change the sound from good to better. It just make it difference. Only priority/preference/taste that can decide which one to pick. I myself, I think I prefer no zobel because it is more smooth, easier to listen to.

Imagine you have to play bass, and you need to strum two strings consecutively at certain speed. You should have a feel regarding how much pressure you should give when strumming the first string, shouldn't you?

Now imagine situation where you strum the first string with lower pressure than you should. The sound will be different, probably not what you want, but it can't be disturbing.

Now imagine situation where you strum the first string with higher pressure than you should. The sound will be different, probably not what you want, but it can be disturbing! Because the transition is not smooth. May be because the first string is still vibrating when you strum the second one and their combined vibration is not in tune. Whatever the mechanism, I believe that too much pressure can be more disturbing than too little pressure. Tho the sound will be more impressive with higher pressure.

May be the above explanation can be the reason why I feel more relax listening to no-zobel. And in real life, I have always prefer subjective performance than objective performance. Take for example the use of zobel at the output solid state amplifiers. Even if it objectively can help against oscillation and easy/linear drive of the speaker due to flatter impedance, I prefer not to use it.

On second thought, I think the zobel will reduce sibilance with TDA. But overall, I believe it cannot be reduced to zero, because I believe it is the switching frequency artifacts. If there is option to reduce gain then reduce it to minimum. Also use lower "sampling" frequency, around 400kHz may be.
 
Last edited: