Tpa3255 running in BD modulation at 600khz.

[Hayk]

From the precedent thread, I understood that the TPA3255 has a design flaw when programed as BTL or PBTL. The sound becomes normal when used as 4xSE.
With 4xSE, channels A and C have same the phase also B and D. This feature permits to use the BD modulation.
I also remarked that this chip has variable dead time, similar to STmicro's STA516, adapting according to the intensity of the current. This feature makes the chip best suitable for BD modulation, eliminating the distortion due to dead time, except at very low power in10-20mw range that can be neglected.
The amp will run at 600khz carrier frequency, but in BD mod, the frequency doubles and the residual will become 1.2Mhz. This means lower resonant Q, higher frequency response, and more post filter feedback.
I will design using LCSC electronic parts to be manufactured by JLCPCB as much as possible.
To avoid noise from 12v step down converter, it will be an external ready made module.
Below is the starting schematic, I need to decide whether the chip should be cooled by a Fan+heatsink or by the casing with the chip on the bottom side.
I will modify my actual board to test. It will use 4.7uH inductors or Sagami type 2x10uH paralleled.

 

[Hayk]

Happy New Year.
It did work. The carrier is 611khz to get doubled at the output filter. The sound is nice at low levels with origin capacitors and opamps.II put the CM choke at the input to get eventually a pre filter feedback for high frequencies.

 

[Hayk]

Which of the 2 inductors to choose.
It needs 4.7-5uH inductors. The TDK PQ2615 is given for 40A rms with thick flat wires and cost only a dollar but it is big 22mm wide.
A compact one is Sagami 7W14A. 10uH x2. It has constant inductance up to 6A but paralleled, is good for 12A. It costs about $7 replicas. I bought once upon a time, equivalent Denon ones a pair, and the vendor sent one fake for free, that costs these days $0.5, for comparison.

 

[Hayk]

This is the simulation of PFFB. It applies 7.5db extra NFB vs 5.5db. The graphs are response for 4ohm, 8ohm and unloaded.
The trouble is that it needs 5vp for maximum output. This means, the input stage with 12v supply must use rail to rail output opamps.

 

[Hayk]

I was enjoying the beautiful sound using my 4 transistors interface, with 3110 and 3255. Today I put back the opamp, after few hours,I got bored. The trouble with simple circuits, you can't get 0.000x THD.
I will try another kind of circuit before I start with PFFB, using LM4562 shown below.
I will use external power supply about 17v and lm317 regulator.

 

[Hayk]

I made it work, see the circuit below. It sounds very nice, more open, more live than the follower+inverter. I need to listen tomorrow several hours before I apply the PFFB circuit, ready to be wired.

 

[Hayk]

Tha bass was very weak pronounced. This is due to power supply rigidity. I replaced the LM317L with TL431+BD139. It has 0.01 ohm impedance and I have full bass, see the updated circuit bellow. I will listen a few hours before applying the PFFB that should not harm the high quality I got now.

 

[Hayk]

To understand what means high resolution amplifier, listen to the 2 pieces of human masterpieces. The Mozart's requiem needs an orchestra of about 70 instruments and 70 vocals, that of Verdi, 300 vocals.
These 2 versions of Karajan and Toscanini, I always thought, they are bad recorded with low performance microphones of the date.
This amplifier proved me, I was wrong. Not only I am distinguishing individual
voices but the female voices are so celestial that I am encouraged to high power levels.

 

[Hayk]

Is this TPA3255 amplifier PFFB?
Oh YES, it IS.
Is it stable?
No oscillations.
I'll tune for square wave, after I validate the sound quality.

 

[Hayk]

After half an hour listening, the only thing I remarked is silghtlly better bass punch.
The 3255 has 2 pins on the A and D outputs but singles on B and C. These tiny leads are flat soldered on a surface still thinner. Maybe the 7.5db feedback is decreasing the effect. The 3110 in PBTL has 2 pins on each branch and thicker than 3255.

 

[Hayk]

This how the circuit looks like so far.

 

[Hayk]

These are the square wave 1khz response with PFFB, without input bandwidth limiter, for
1 No load
2 8ohms
3 4ohms
With 1k and 220pf input filter, the leading spike can be eliminated.

 

[Hayk]

The frequency response with 8 ohms without output snuber, peaks 3.5db at 195khz, after 0db flat to 90khz.
With 4 ohms, it is flat to 22khz and reaches -3db at 125khz without peaking.

 

[KL....]

Thank you @Hayk, you might like to look at the following....

• Neurochrome Universal buffer Neurochrome Buffer Spec

 

[MCH]

After half an hour listening, the only thing I remarked is silghtlly better bass punch.
The 3255 has 2 pins on the A and D outputs but singles on B and C. These tiny leads are flat soldered on a surface still thinner. Maybe the 7.5db feedback is decreasing the effect. The 3110 in PBTL has 2 pins on each branch and thicker than 3255.

Don't worry about the size of the pins, the wires inside are much thinner, but it has more than one per mosfet (still thinner combined than the pin).
To your point, i believe the number of wires for A and D is 5 and for B and C is 3. not 100% sure though, if i find the time i will try to find or take a better picture:

(you can get an idea of the diameter of the wires used looking at the one still hanging out at the right side, on top of the shadow)

PS: could you please post graphs of the frequency responses you get? thanks!!

 

[Hayk]

Thank you for posting the picture of the chip.
The wires used to link the chip to pins are gold to be welded by ultrasound friction. If you look the schematic, each output is accompanied by its power supply and ground to adjacent pins. The reason for this is shown on the AWG table. This table gives the value of maximum current for power transmission, obeyed "religiously" by transformer-coil winders, and the maximum current if used in chassis, magnetic metal. When the magnetic field gets constrained in the wire, the maximum current gets considerably increased. To constrain the magnetic field, paralleling inverse current carrying wire, or using magnetic wires, can also work. Capacitors, nearly all, that need to pass high pulse current use magnetic leads as well as high current resistors.
The problem with SMD components as 3255, is the low temperature solder holding high current leads. The surface is too small to pass 8A rms.
The PCB needs 2oz/ft cupper, hence higher cost, for only high current tracks. A cheaper solution is to leave these tracks bare and fill with 10% silver solder.
Better solution that also avoids electromagnetic radiation, is to add on the bare tracks magnetic wires, either capacitor leads or acoustic guitar strings, brass wound steel.

 

[MCH]

they are actually copper, but doesnt really matter. The point is that not all the 4 channels are wired the same for some reason, which i find curious.

 

[Hayk]

Can you check if they are magnetic. Rub a screw driver on a magnet and see if it does act on the wires.

 

[MCH]

Can you check if they are magnetic. Rub a screw driver on a magnet and see if it does act on the wires.

Need to find the die first, I decapped it last year
But if I find it, will do, no problem

 

[Hayk]

The TPA3255 costs $3.5 from LCSC where JLCPCB will buy and the price will continue fall in near future. I will use 2 chips, one for each channel to use a passive heatsink shown bellow that will be installed horizontally inside the chassis, to be cooled through bottom and top holes.
The circuit will be on 3 PCBs, a vertical that carries the 3255s, a bottom horizontal for the filters with PQ2615x4 and a top horizontal for the input interface.
It makes a compact 75x40x40mm module + 43mm wide for the heatsink.