So the TPA325x series of chip amps came out about a decade ago, yes really that long! My datasheets for the TPA3251D2 are dated June 2015 and that's where we happen to be now only 10 years later, so, happy birthday TPA325x series. 
🥳🎂
Over the years I've seen it mentioned a couple of times that one could go composite with the TPA325x using an ncore style integrator only I'd never seen it actually done. Now my main amplifiers, for my active speakers, have been based on a PFFB implementation of the TPA3251. The performance is more than acceptable but I'd always wanted to try and go that bit further so figured I'd give the integrator a go. It wasn't my plan for this to coincide with the chips ten year anniversary but there you go.
As you can imagine it took a considerable amount of trial and error, plus simulation time, to get things working correctly. This is especially so as there are no models provided for any of the TPA325x series so everything required a large dose of guestimation.
4 layer PCB from China with the TPA mounted on the bottom side of the board. The PCB being bolted to a bar of aluminium to get the heat out. The bar then being bolted to a case panel once 6 channels have been built.
So some performance graphs.
First up frequency response into no load, 9.4 ohms, 4.7 ohms, and 2.35 ohms.
Next as ASR has made popular 1kHz 5 watts into 4.7 ohm load 48kHz bandwidth unweighted.
Next the same thing but 51 watts.
Here's a two tone 18500Hz + 19500Hz, as Bruno liked to show off with the Ncore amps, at 5 watts into the 4.7 ohm load.
And now at 51 watts.
Next up is a distortion sweep at 5 watts into the 4.7 ohm load 96kHz bandwidth just so we can see what the 3rd harmonic really does towards 10kHz.
And finally a distortion sweep at 51 watts 96kHz bandwidth.
Anyone who is familiar with the TPA325x will immediately notice the significant improvement in high frequency distortion. The main area the TPA325x has issues is at the high frequencies. I'm assuming that, for stability, TI needed to start compensating the amplifier pretty low down in frequency as can be seen from distortion Vs frequency sweeps of the standard implementation.
Here's the datasheet distortion at 80kHz bandwidth.
At higher power levels it can be seen that the distortion starts to rise as low as 100Hz! The composite version completely does away with that and reduces the overall magnitude significantly. So happy 10th year TPA325x and enjoy some higher performance.
Now I wonder when Chi-Fi will come up with something similar. Complete schematics not shown for obvious reasons (at least yet). If China want to copy it they are welcome to, and possibly improve upon it, however they will at least have to figure it out for themselves! Just knowing it can be done should be enough encouragement for them to try. Topping are you there? 🤣
I need to build 4 more channels and then I finally get to listen to something.
By the way TI hurry up and come out with a replacement for the TPA325x series using GaN FETs. You've already done an integrated GaN motor driver gimme some class D. I want the improvement in efficiency and idle power consumption. Not to mention that the ability to switch faster should reduce the needed deadtime and improve performance.
As to the clipping of the TPA325x composite. The usual ncore clipping detector and limiter is used although I haven't tried to really fine tune things. Nothing, so far, has blown up. The TPA325x amps are very good at shutting themselves down if anything untoward happens, and that happened a lot when trying to get this stable. The window of stability being quite narrow. Not enough feedback and you oscillate, too much feedback and you also oscillate. You need to find the Goldilocks zone of stability. One nice thing about the TPA chips is they have a clipping detector so should it be necessary you can shut the amplifier down at the onset of clipping.
🥳🎂Over the years I've seen it mentioned a couple of times that one could go composite with the TPA325x using an ncore style integrator only I'd never seen it actually done. Now my main amplifiers, for my active speakers, have been based on a PFFB implementation of the TPA3251. The performance is more than acceptable but I'd always wanted to try and go that bit further so figured I'd give the integrator a go. It wasn't my plan for this to coincide with the chips ten year anniversary but there you go.
As you can imagine it took a considerable amount of trial and error, plus simulation time, to get things working correctly. This is especially so as there are no models provided for any of the TPA325x series so everything required a large dose of guestimation.
4 layer PCB from China with the TPA mounted on the bottom side of the board. The PCB being bolted to a bar of aluminium to get the heat out. The bar then being bolted to a case panel once 6 channels have been built.
So some performance graphs.
First up frequency response into no load, 9.4 ohms, 4.7 ohms, and 2.35 ohms.
Next as ASR has made popular 1kHz 5 watts into 4.7 ohm load 48kHz bandwidth unweighted.
Next the same thing but 51 watts.
Here's a two tone 18500Hz + 19500Hz, as Bruno liked to show off with the Ncore amps, at 5 watts into the 4.7 ohm load.
And now at 51 watts.
Next up is a distortion sweep at 5 watts into the 4.7 ohm load 96kHz bandwidth just so we can see what the 3rd harmonic really does towards 10kHz.
And finally a distortion sweep at 51 watts 96kHz bandwidth.
Anyone who is familiar with the TPA325x will immediately notice the significant improvement in high frequency distortion. The main area the TPA325x has issues is at the high frequencies. I'm assuming that, for stability, TI needed to start compensating the amplifier pretty low down in frequency as can be seen from distortion Vs frequency sweeps of the standard implementation.
Here's the datasheet distortion at 80kHz bandwidth.
At higher power levels it can be seen that the distortion starts to rise as low as 100Hz! The composite version completely does away with that and reduces the overall magnitude significantly. So happy 10th year TPA325x and enjoy some higher performance.
Now I wonder when Chi-Fi will come up with something similar. Complete schematics not shown for obvious reasons (at least yet). If China want to copy it they are welcome to, and possibly improve upon it, however they will at least have to figure it out for themselves! Just knowing it can be done should be enough encouragement for them to try. Topping are you there? 🤣
I need to build 4 more channels and then I finally get to listen to something.
By the way TI hurry up and come out with a replacement for the TPA325x series using GaN FETs. You've already done an integrated GaN motor driver gimme some class D. I want the improvement in efficiency and idle power consumption. Not to mention that the ability to switch faster should reduce the needed deadtime and improve performance.
As to the clipping of the TPA325x composite. The usual ncore clipping detector and limiter is used although I haven't tried to really fine tune things. Nothing, so far, has blown up. The TPA325x amps are very good at shutting themselves down if anything untoward happens, and that happened a lot when trying to get this stable. The window of stability being quite narrow. Not enough feedback and you oscillate, too much feedback and you also oscillate. You need to find the Goldilocks zone of stability. One nice thing about the TPA chips is they have a clipping detector so should it be necessary you can shut the amplifier down at the onset of clipping.
Really impressive values. Good job, congratulations.
Last ten years I have made many cards with TPA325xs . None of them were with PFFB. And I did not measure for any of them; But each of them has a background noise, especially in "hiss" style and high frequency. Since I could not solve it, I retired them all in a way.
However, I have recently made a PFFB project and PCBs have yet reached my hand. The PFFB line consists of only 18k single resistor. I also obtained the opamp supply from low noise symmetric regulators (+/-12v) , and I received the analog part of the TPA from the same symmetrical supplies + rail. I haven't installed yet, I hope I can get close results as yours.
Regards ..
By the way; which opamps you used for this project?
Last ten years I have made many cards with TPA325xs . None of them were with PFFB. And I did not measure for any of them; But each of them has a background noise, especially in "hiss" style and high frequency. Since I could not solve it, I retired them all in a way.
However, I have recently made a PFFB project and PCBs have yet reached my hand. The PFFB line consists of only 18k single resistor. I also obtained the opamp supply from low noise symmetric regulators (+/-12v) , and I received the analog part of the TPA from the same symmetrical supplies + rail. I haven't installed yet, I hope I can get close results as yours.
Regards ..
By the way; which opamps you used for this project?
I'm surprised that the amplifiers had an appreciable amount of hiss usually the TPA325xs are pretty quiet. With PFFB they are even quieter. If you were using compression drivers of 110dB+ sensitivity, in a home system, then you might be able to hear something.
The amplifiers are pretty tolerant of the power supply that you're using too especially with regards to hiss.
The low voltage supplies on the TPA aren't critical to supply quality either. The gate driver supplies just need to have enough local decoupling to be able to switch the low side FETs quickly enough and the high side get their charge from the boot strap caps. Internally the analogue section has its own regulator too.
Are you following the TI application note for the implementation of the PFFB? A single resistor won't typically do the job.
I've used LME49724, OPA1632, and OPA1633. The OPA1632/33 are interchangeable but I figured I'd give the newer version a go just to see what it's like. Note that my design isn't a standard PFFB implementation like TI describes as it places an opamp integrator inside the PFFB loop to provide additional feedback.
The amplifiers are pretty tolerant of the power supply that you're using too especially with regards to hiss.
The low voltage supplies on the TPA aren't critical to supply quality either. The gate driver supplies just need to have enough local decoupling to be able to switch the low side FETs quickly enough and the high side get their charge from the boot strap caps. Internally the analogue section has its own regulator too.
Are you following the TI application note for the implementation of the PFFB? A single resistor won't typically do the job.
I've used LME49724, OPA1632, and OPA1633. The OPA1632/33 are interchangeable but I figured I'd give the newer version a go just to see what it's like. Note that my design isn't a standard PFFB implementation like TI describes as it places an opamp integrator inside the PFFB loop to provide additional feedback.
In fact, TI has two version of PFFB application notes of TPA325x series: the first one is SLAA702 and in this application note, TI uses only one 18K resistor. But the second version SLAA788A is different and they added two caps and one resistor to the GND.
If I remember correctly the 18k only version used a more aggressive zobel on the output filter to stabilise the amplifier. Whereas the second iteration uses a cheaper zobel but adds phase compensation in parallel with the feedback resistor instead.
No this is not a TI PFFB implementation. As I mentioned in the original post this is a composite amplifier with an opamp based integrator within the outer feedback loop. This is akin to what Bruno Putzeys did with the ncore amplifier for it's performance gains Vs the standard UcD amps.
TIs PFFB is entirely passive putting a bit more feedback into the TPA325x itself. It doesn't introduce another active device into the feedback loop. The opamp they use is there just to ensure a low source impedance for the PFFB network.
TIs PFFB is entirely passive putting a bit more feedback into the TPA325x itself. It doesn't introduce another active device into the feedback loop. The opamp they use is there just to ensure a low source impedance for the PFFB network.
Nice.
I believe Topping already have one. It's the PA5 using TPA3251.
THD+N at 0.0005%.
There's a lot of issue with this model and being discontinue. Replaced with a Mark 2.
I believe Topping already have one. It's the PA5 using TPA3251.
THD+N at 0.0005%.
There's a lot of issue with this model and being discontinue. Replaced with a Mark 2.
The PA5 doesn't use an integrator. It's just a typical TI PFFB implementation with different component values. You can find schematics for the feedback circuit online.
You're right.
https://www.audiosciencereview.com/...ix-d01-module-replacement-for-everyone.44219/
Did not realized someone actually reverse engineer it 🙂
https://www.audiosciencereview.com/...ix-d01-module-replacement-for-everyone.44219/
Did not realized someone actually reverse engineer it 🙂
Congrats for your achievement. Without disclosing proprietary information, would you be able to say if this effort required going beyond the standard linear system identification and compensation methodology? It sounds like you made a significant effort to reverse engineer the TPA325x. Also, in general, how sensitive is the amp to load impedance? Thanks, I’ve been thinking about trying to do a composite TPA325x amp for a while.
Impressive work, congrats. Most interesting to me is the linear and nearly load independent frequency response.
Smells like Bruno😎
Smells like Bruno😎
@5th element Absolutely ridiculous performance, I hope someone accepts the challenge!
Now, I am willing to accept something very slightly less stellar like 110dB/THD+N, but I want more power pretty please.
TI needs to make a 400W/channel in 8 Ohm version, getting spoilt for choice in the <50W region, 100W is only 3dB more and if I want 100W with 6dB overhead that is 400W... 😢
The two things that were the most help in achieving success in this design were fully understanding the concept behind the typical Ncore amplifier. We're very lucky that this paper happens to exist.
https://www.semanticscholar.org/pap...Kemp/0b7d5b001c98bd27650ff9fb55d44bd47b29dd8d
Everything is laid out there and although my maths isn't up to the job of understanding all the complex stuff it does a fine job of describing how the additional integrator control loop is designed to work. It's actually very elegant!
The second thing was seeing a talk that Bruno made on the Ncore...
...where he's discussing the overall gain and margins of the device. That helped to connect a few dots as well in terms of what to shoot for when applying compensation.
If you've got the amplifier compensated correctly then it is very tolerant of load impedance changes. You can see the invariance in the measurements I posted above. However the amplifier is at its least stable when driving an open load. Load it and the phase margins increase significantly this is useful because we do tend to use our amplifiers with things connected to them 😀
With regards to reverse engineering the TPA325x I did my best to come up with some sort of facsimile so that I could hopefully get simulations that would be good enough to help with the compensation and then give a direction for fine tuning it. The thing that helped the most was that I made the assumption that all class D amplifiers, that take their feedback point after the output filter, are intrinsically bound, in terms of bandwidth, by the filter network itself. Given that these filter networks all happen to roll off at roughly the same frequency, and are second order, I assumed that the filter network would dominate Vs whatever compensation is used internal to the TPA325x. The Ncore concept relies on the fact that the basic amplifier, it's applied to, is unconditionally stable. I figured that if a TPA325x is unconditionally stable, plus the output filter placing an upper bound on bandwidth, that it should hopefully be possible.
There's nothing magical to the compensation. I just used the basics I've picked up on over the years and then saw what sticks.
Well I definitely stood on the shoulders of giants in getting this to work!
We will have to see what TI come up with next. The trouble is parameter optimisation within the chip amps. The first major limitation is the size of the chips. They are very small and thus have limited die area to dedicate to the output MOSFETs and for getting rid of heat.
You want to keep the RDSon of the MOSFETs low so as to minimise power dissipation under heavy load. Lower RDSon requires bigger MOSFETs and we're limited on space. Increase the size of the FETs and they become harder to drive. This increases the idle power consumption and reduces the total thermal capacity of the tiny chip for maximum power outputs. This is probably one of the reasons TI went with a lower gate drive voltage in the TPA3255 Vs the lower power versions.
If you want to increase the voltage that a FET can work at this also requires that you increase the size of the device. Oh dear.
Now the one saving grace of higher impedance loudspeakers is that they do not need as much current as low impedance ones. This means that you could go with 100V operation, the FETS RDSon would go up, but it wouldn't be too much of a concern because then you're only driving 8 ohm loads, have minimal current demands, and the I2R losses would go down. The chip amp would operate efficiently when driving 8 ohm loads and would still come with robust over current and over temperature protection built in. The caveat being that if you did try and drive lower impedance loads that the chip would shut down if it got too hot. But it shouldn't blow up.
You can understand why TI chose to optimise things around lower than 8 ohms.
Interestingly though Infineon/International Rectifier did just what you are asking for with their MA5342MS chip. A higher RDSon version that can operate at higher voltages. You'll get more power into high ohms but it's ability to drive low ohms is poorer Vs their other MA5332MS.
GaN transistors would provide output devices with lower RDSon and lower gate drive characteristics, for a given surface area, for a given voltage Vs silicon devices. If TI were to make hybrid GaN/Silicon TPA325x replacements they could theoretically do it with 100V devices but would they go that route?
And then there were six channels!
They sound as good as the measurements would suggest. The most striking thing is how incredibly quiet they are. The six channels are driving these speakers actively.
https://5een.co.uk/FST Cardioid Concept.htm
The B&W FST has a sensitivity of 95-100dB across its bandwidth and with ear right up against the cone you hear nothing. The TPA3251 with PFFB you could still hear a small amount of residual hiss.
They sound as good as the measurements would suggest. The most striking thing is how incredibly quiet they are. The six channels are driving these speakers actively.
https://5een.co.uk/FST Cardioid Concept.htm
The B&W FST has a sensitivity of 95-100dB across its bandwidth and with ear right up against the cone you hear nothing. The TPA3251 with PFFB you could still hear a small amount of residual hiss.
This invariance is a great achievement imho. Keep in mind that real speakers do not provide something like real 8Ohms at resonant frequency of the output filter. For this reason I suggest compensating for open load.