Advance warning: this is a long read. If this subject interests you, get yourself a coffee now, or something stronger.
PREFACE
In 2010 I assembled a Tripath TA2020 amp board - the diyparadise Charlize 2, which was at that time considered the best of the TA2020 boards, and the TA2020 was still flavour-of-the-month as budget hifi amplifier.
For power supply I chose the well regarded Paul Hynes PR3G2 13.8V regulator, fed from a conventional 120VA EI transformer.
Then life intervened, and it took me another 7 years until I eventually built up a music server/DAC plus Planet10/MarkAudio speakers, to complete my hifi setup. It took another 6 months to sort out a few earthing/ground loop issues, and for the speakers to fully break-in. I am now reasonably happy with the sound, but the bass is a little over-emphasised, like when a loudness knob is turned up. Also I can't help thinking that the high-end nature of my transport, DAC, and speakers makes my system deserve an amplifier which is more than just "budget champion" status. Of course I've heard the hype about the new budget class D amplifiers such as the TPA3110/3116/3118, and the more recent TPA3255, but my preferred upgrade path is (or was) to leapfrog class D altogether, and go for a class AB amplifier such as the Ranchu/AKSA Quasi-Complementary, or the Apex AX11.
Meanwhile I became interested in building a small boombox-style music system in order to hear my music outside my house, on the deck, without needing to crank up the main system to carry the sound outside.
So I recently decided to buy and test the well regarded (and cheap) Sanwu TPA3118 and Sanwu TDA8932 amplifier boards. I had read the long forum posts regarding improvements to these boards, and I was concerned that my TA2020 was built to a higher standard, thus it wasn't going to be a fair comparison unless I did some basic upgrades to the Sanwu boards. So this is what I did:
- replace the output inductors with Wurth WE-PD 22uH inductors.
- bypass the onboard input caps, then wire IN+, IN-, and GND to 3pin XLR connectors. For some tests, good quality input capacitors (Wima MKP10) will then be wired outboard.
- for the 3118 - gain reduction mod, by removing R27.
- for the 8932 - add more capacitance to pin 12, in preparation for full balanced connection from DAC output transformer.
Power supply will be a standard 19V SMPS, smoothed with a capacitance multiplier from ALLO. The cap multiplier is set to drop 0.1V.
And let me say that these small Sanwu boards are rather hard to work on!
SOURCE AND SPEAKERS
My transport is a BeagleBone Black with Twisted Pear Cronus isolation/reclocking board, mounted directly atop a QuangHao ES9018 DAC with Lundahl LL1684 transformer output stage. This output stage can be configured for unbalanced output, or balanced output, via jumpers. Since there's no DC offset from a transformer, it's possible to bypass input capacitors on the amplifier.
My speakers are Alpair 7 gen3 (fullrange) in Planet10-designed onken-style trapezoidal cabinets, which are designated "dMar-Ken7.3mT".
Since the purpose of these tests was to choose the best amp for a cheaper music system, I thought it would be a good idea to also test with a cheap speaker - and I already have a good cheap speaker lying around - the Micca MB42X, which is sold here in Australia as the "Voll B44".
Test outcomes were the same using either MarkAudio or Micca speakers, but the Miccas revealed differences more obviously.
LISTENING TESTS
1. DAC output unbalanced, via input caps.
First I tried the TA2020, since I'm quite familiar with this setup. As I mentioned earlier my only criticism here is that the bass is a little over-emphasised.
Then the 3118 - more detail, clarity, precise treble, and while the bass was still a little over-emphasised, the music was cleaner. I'm now getting an idea of what the TA2020's alleged "tube-like" characteristic might be.
Then the 8932 - similar to the 3118, just not as refined, and some harshness in the treble.
2. DAC output unbalanced, no input caps.
TA2020 - I could maybe detect just a very subtle improvement. The Charlize2 amp is supplied without input caps, because the designer claims it sounds better this way ... but I must say the difference is barely discernible.
3118 - a very small improvement in clarity. Bear in mind that the input caps in the previous test were quite good quality, and they don't seem to impede SQ by much.
8932 - does not work in this configuration.
3. DAC output balanced, secondary centre-tap to GND, no input caps.
3118 - ah, the over-emphasised bass is now gone, with a nicer tonal balance, and a clean, balanced sound.
8932 - does not work in this configuration.
TA2020 supports unbalanced inputs only.
4. for 8932 ONLY - unique connection from secondary centre-tap to amplifier pin 12 (input reference voltage)
Similar to previous result with 3118 - nice clean tonal balance - but occasional harshness in the treble remains.
RESULTS, CONCLUSIONS
I consider myself a rational sceptic, and I was never convinced by the hyperbole surrounding the TPA3118 and TDA8932 - specifically that their SQ eclipsed all that came before, including the TA2020.
But now I'm happy to concede that they are, indeed, better than the TA2020.
The 8932 is definitely in 2nd place behind the 3118. There are comments on this forum that the Sanwu 8932 amp board can be improved by cleaning up the voltage applied to the analogue supply pin, so that its SQ then leapfrogs the 3118, but I can say that presently the Sanwu 3118 sounds nicer than the Sanwu 8932 board.
UPDATE: just today I see an optimised TDA8932 board emerge on the forum! It will be interesting to see how it performs.
Balanced configuration with the 3118 is a clear winner in these listening tests ... but wait ... I don't suggest everyone with a 3118 amp should rush out to get a balanced DAC! This result should be seen in light of my particular situation - I have a transformer output stage with a specific configuration. Sure, you might expect these class D amps to sound slightly better using balanced input compared to using unbalanced input, but I don't imagine the difference to be as stark as the difference that I heard. No, I think this difference is more due to my DAC transformer output stage - I think its performance/characteristics are starkly different between its balanced and unbalanced configurations. This may be related to the topology - my transformer has a filter network on the primary side tailored to suit the ES9018 DAC.
On a related note, over in the TPA3118 forum thread I see it's popular to use an input transformer in unbal-to-bal configuration instead of input capacitors. This is similar to my situation, but not the same - especially in terms of the aforementioned filter network of my transformer. So it makes me wonder which aspect(s) of this input transformer achieve the apparent improvement in SQ:
i) using the amp in balanced input mode instead of unbalanced input mode?
ii) eliminating input caps? (in my tests, good quality input caps don't hurt the SQ of the 3118 very much)
iii) adding a transformer flavour to the sound?
iv) a combination of the above?
WHAT NEXT?
To get a better idea of the different quality outcomes of using unbalanced or balanced inputs of a TPA 3118, it would be useful to employ some form of output stage OTHER than a transformer; say a solid state device with switchable unbalanced/balanced outputs. Or maybe the direct output from a PCM5xxx DAC chip. The PCM5122 is unbalanced, and the PCM5242 is balanced. And both have no DC offset, so you can go cap-less into the amp.
But I've lost interest in any further tests at this stage. Winter is closing in, and building a 2nd system/boombox has fallen down the priority list!
For now I think I will probably swap out the Charlize TA2020 amp in my main system with the 3118 amp - and obviously connect the new amp in balanced configuration.
Can I be bothered to box up the 3118 boards in a nice chassis? Probably not. This generally ends up taking more time than you expect, and also I need to take some time to consider my next step with the hifi system overall. I now know that my source device is (significantly) better in balanced mode, so the unbalanced class AB amps I was previously considering are now probably off the list. If I stick with the TPA3118, maybe I should get one of the "premium" boards such as the newly released one from ALLO? Then again, current flavour-of-the-month seems to be the TPA3255, or for my fullrange speakers maybe more suitable would be its lower power brother the TPA3250?
And then there's the TPA3244 - I don't think anyone here has heard this amp yet?
Must sleep on it.
