I’ve heard people saying that Ti’s TPA3116 class-D amplifier sounds better than Tripath’s TA-2020 class-T amp, so I bought one to try out: just US$8 on Taobao. There are a lot of different ones available, but none with air-core output inductors, which I know I prefer to iron/ferrite cores. However, most amps have the same low grade output coils and this amp has toroids and the core material is coloured red signifying a higher grade than the usual yellow core. In addition, I read the datasheet and looked for one designed as closely to that as possible. Here it is, the “big” blue Sanwu version, slightly more expensive than the red one or “little” blue one, with apparently better components than either:
It arrived and is almost exactly as pictured, plus under the output coils are several smd resistors and capacitors too – they are the C-RC output filters specified in the datasheet – I guess I’ll need to add the bootstrap RC from the EVM datasheet. One disappointment is the ground plane – a grid of copper traces rather than a solid sheet, which might limit return currents and dynamics. The bootstrap caps are also quite a way from the IC so that creates a long feedback loop – not ideal for high frequency, high power pulses. Finally, while hooking it up, my finger went onto a corner of the heatsink and it came off – weak glue – but this is no biggie – I wanted to measure the chip temperature and check component values anyway.
So it’s hooked up and burning in the caps, which I don’t expect will take long. I’m running it from an ES9023 dac and powered by the excellent Sigma11 psu I built for the Kef Egg TA2020 amps. I plan to do a back to back listening test comparing this amp with the SKN TA2020 amp, but initial impressions are very good. The amp doesn’t pop when powering on (Psu has a soft start) or off (lots of onboard capacitance) and the sound doesn’t appear to have any obvious deficiencies either. At “loud conversation” sound levels (~85dB ish), the chip stays cool with no heatsink – about 5-10 degrees C above the room temperature in free air. I cleaned off the gunk to check it is a TPA3116 too – lots of fakes on Taobao – and at least it says it is one on the outside.
After buying the board on a whim and hunting for more info, I found the wiki on diyAudio so I have a decent idea about what to do if I want to mod this amp, however, I don’t know what value the output coils are so I’m not sure if they’ll suit 4, 6 or 8 ohm speakers. Based on the capacitor values and that all the components are the same value as the datasheet, I assume the output filter is already set for 4 ohm and my speakers are 4 ohm woofers with 6 ohm tweeters. It does not sound over-sparkly and that’s what matters.
After a day burning in, I hooked up both amps so they get the same signal each and one speaker each in order to do an instant switch over and compare their character. I carefuly set the volumes to match and listened. It’s immediately clear they are not hugely different, but after listening for about an hour, the different tonal balance becomes equally obvious. The TA-2020 is noticeably tipped towards a warmer, meatier sound that also makes it a little muddier, with a softer treble and more prominent vocals. The TPA3116 has a brighter sound with less prominent bass, a slightly crispy aggressive treble, and clearer vocals. Which one is better? I think that will depend on what speaker you pair it with and how much either can be improved – certainly I would not use this amp to drive a revealing tweeter like the Klipsch have, but it’s fine for a silk soft dome. The TPA is almost silent and the TA2020 has a tiny amount more of background hiss and both chips will thump on without a soft start psu, but the SKN doesn’t because it has delayed relays on the output.
The best first step is to reduce noise, and that is layout, set up such as gain, and power. I am using an excellent psu and can’t change the layout much so the first mod was set up: lower the gain from 26dB to 20dB by removing a 100k resistor (R27) and paralleling another (R28, 20k) with a 10K so it is 6.6K. The datasheet suggests 5.6k but with R27 open, an exact value is not critical to set the gain at startup. Lower gain stops the amp going into heavy distortion at full volume for my set up (12 V psu give 20V output swing from 2V input – a gain of 10 or 20dB), and it also raises the input impedance to 60k so the 1uF input caps don’t affect very deep bass – well below the speaker’s range in fact. Listening showed this change only had one noticeable effect – less gain, so I switched back to 26dB. I had hoped the extremely quiet background hiss would fully disappear but it did not.
With noise done, the next logical step to improved performance is the output so I tried different bootstrap caps (above). The datasheet recommends X5R ceramic but X7R are better. They both have more distortion than NPO or metallized film caps. So I tried 100v 220nF metallized poylpropylene on one channel to compare with ceramic on the other. The poly caps sound softer and more natural so I fitted them to both channels.
Next mod was the bootstrap snubber – not in the datasheet but used in Ti’s EVM – these are typically 10-20 ohm resistors with a small NPO cap around 330pf. The datasheet says 10r and 330pf but without knowing the PCB layout, this is also an estimate. I can’t do the measurements/calculations needed to know exact values so I chose 15R and 270pf. I fitted them to one channel as a return path to the chip. The sound was a little more detailed and defined, with a little less of the aggressive crispiness. It’s subtle, but better, so again, fitted to both channels.
Finally, the last thing is to cap swap (signal and power) to change the tonal balance and sound quality, especially to gain a weightier bass. That means larger signal input coupling caps (unlikely to have much effect as they are simply DC blocking high pass – more likely to change treble detail and distortion) and putting larger power caps nearer the chip – this should change the character. The caps on the board are solid polymer which, when I have tried them in other projects, can seem a little over-bright. They have a very low ESR and are good for filtering SMPS noise though, and can handle large ripple currents – there are eight of them on the board with very wide traces to the IC. So I plan to try an audio grade cap right next to the IC – Elna Silmic II (quite dynamic and my default choice) and/or Panasonic AMX (a neutral, natural sound and more laid-back than the SilmicII – might suit this amp).
Cut a long story short – the caps changed the sound, but I didn’t feel any were an improvement – generally they added detail to the treble, which boosted sibilance, which I didn’t want. And I didn’t manage to add weight to the bass, which was the aim. In fact, the original caps are fine and suit this amp quite well so, like the gain, I just reversed these changes. Next up was putting it in a case: an old Musiland MD-10 DAC case with a disused PGA2311 remote control volume. It took quite a lot of hacking and drilling and cutting and gluing but it fits. Together, it’s a tight fit but works well and looks good to me – polishing the front took a long time but it is a good match for a paint I found – Pylox 334 Midnight Blue.
The amp runs from a 12V 2.5A smps psu that came with a Buffalo NAS, but this creates a switch on thump so next up is to modify the mute circuit to provide a delay. This is simple as described here. A schottky diode in parallel with R30, and the chip-side of R30 and R23 are connected by 150K + 4.7uF in series. I didn’t have those values but saw that using anything over 100k and doubling the cap to 10uF would work, so I used 130k and 10uF. It didn’t work (silence even after a minute) so I checked everything many times again and guessed that the cap must be taking too long to charge so the amp stays in mute. I tried a few combinations and eventually got 100k and 1uF working – and the amp is mute at power on just long enough to avoid a thump, with only a tiny click a few seconds after powering off. Sweet. I tried to leave some room for a heatsink if I need one, and ran the cap lead perpendicular to the traces to minimise interference.
After that, I added a bass boost circuit after the PGA – it’s passive so it will attenuate the signal by 6dB going into the amp because in order to simulate a bass boost, it is actually an attenuation of everything but bass. There’s still enough gain so this is not a problem. This should make the amp more meaty sounding and suit the bookshelf speakers. It’s easy to do: a 10k variable resistor in series and another 10k VAR + 1uF shunt to ground. Set both VARs to 5k to attenuate -3.5 dB at 32Hz, -4.5dB at 63Hz, -5dB at 125Hz, -5.5dB at 250Hz, -5.75dB at 500Hz, -5.9dB at 1khz and so on, such that changes in attenuation are barely audible above 250Hz and inaudible above 1khz. After a little experimentation, I created a shelving bass boost with 3k in series and a shunt: 2.5k in series with 1uF and 5K in parallel. The 5k and 1uF in parallel set the frequency at 32Hz, and the other resistors set the attenuation curve. So at 1khz and above, the cap is almost a short giving 3k in series and 2.5k shunt: max attenuation is -6.8dB. At 32Hz the shunt is 5k so attenuation is -4dB => eq. 2.8dB boost. Maximum boost has the shunt at 7.5k, -2.9dB, so 3.9dB eq. boost. I used Blackgate N 1uF 50V non-polar caps bypassed (paralleled) with Vishay MKP 1837 10nF. I have better caps but these are very small and, with the bypasses, sound like they’re not there. A curious thing happened when I switched the PGA2311PA for a PGA2310P – a higher voltage but lower grade chip: I got a thump at switch off with the 2310 but not with the 2311. Both have the same 0.15uF X7R on the digital supply pins and 0.1uF MKP on the analog supply pins so it is just a chip swap. Anyway, the 2311 works better so I swapped back.
A note about the speakers – they came with an Onkyo FR-V77 and are D-V77A. However, I damaged a woofer so I replaced it with a cheap 130mm driver I pulled out of an Edifier computer subwoofer. I didn’t expect much but this driver had a much deeper response than the Onkyo, albeit much less efficient too, yet sounded great. So I modified the crossover with Solen coils and Sonicaps, attenuated the tweeter by 6dB, and extended the reflex tube. They are capable of surprisingly deep bass for a bookshelf but need a boost low down to do it – the design is basically a -3dB extended shelving. I recently replaced one of the drivers because the foam edge was breaking up – found the same driver is now used in the Edifier M1380. I took out the driver and threw the rest away except for the transformer. It’s an awfully low grade bit of electronics. This TPA3116 amp is intended to replace the Onkyo FR, which is very old – the volume potentiometer track is noisy and the CD no longer loads…
The remote is a learning type – Chunghop L102 which I programmed from the original crappy thing that came with the PGA kit. I also programmed my LG G5 – I love that I can use my smartphone as a universal remote. 🙂