70W stereo 0.005% THD+N amplifier TPA3251D2EVM

In February 2017, Texas Instruments have a special offer on their TPA32xx class-D amplifier evaluation modules – 50% off if you add in TPA32CES as a discount code. So I had to. How could I not with specs like those? So I got two for bi-amping. I ordered each one separately so I could bundle in other EVMs, which they limit to one per order (TPS7A4700EVM and TPS7A30-49EVM) and get some samples of OPA1612 etc. to try. I bought the highest power version of the lowest distortion chips – the TPA3251.  Some say the TPA3244 sounds better in the upper midrange but since I plan to modify the amps from their “test bed” grade, that’s not a factor I considered relevant. The EVM datasheet shows it uses some very very noisy regulators, an oldie-but-goldie op amp, the NE5532, and electrolytic 10uF audio coupling caps, so I expected it would need modifying to get the best out of it. However, straight out of the box, at stock settings, with no time to burn in caps, and from a laptop 19V 5A supply, it sounds good. Deep, clear, smooth …and dead silent… however, 19v doesn’t seem enough or perhaps the gain is too low (edit: it’s fixed at 20dB, i.e. voltage x10, according to the datasheet).


The first psu I tried uses a standard AMB sigma11 regulator built with Elna Silmic II caps, and set to 19V to avoid it dropping out of regulation. The transformer is a Talema 80VA 18ACV-0 toroid capable of 4.2A and providing 26V after the bridge rectifier. Using my phone as a source, (which I assume has a standard line level output of 0.77V rms, 2.2V peak to peak, i.e. 1mW into 600R), using 4 ohm speakers, it will play cleanly at close to full volume but this puts the amplifier into clipping  – an orange LED flashes with bass peaks – even though the sound is not obviously distorted. At full volume from my phone, the amp will also flash a red light and occasionally shut down for 1/2 second.


So I changed the transformer to a 150VA 30ACV-0 transformer – this gives 42V after the bridge rectifier, unloaded, and means I can dial up the regulator output voltage. The amp no longer goes into shut down and the clipping light stops flashing with a regulator voltage of 24V. So a 24V supply allows a 10x gain of 2.2V peak to peak signal. After this, I changed the source to a standard 2v rms output DAC and turned up the volume and voltage again. I can get to within 2dB of full volume with no clipping light, but no matter what voltage I choose, I cannot get the last 2dB without an infrequent flash. However, the sound is still clean and clear and it’s far louder than I ever listen at. P1080123.JPG

I also measured the voltage after the rectifier while playing full volume and it dips to 39V so I have the regulator output set to 34V to be sure it will not drop out of regulation, and this also doesn’t ask the regulator’s pass mosfets to dissipate more than 15watts each so they won’t get hot. Ideally, this amp wants a 250VA 32VAC transformer to power it to its maximum. 3251-voltage-vs-output-power

The next things to consider are the analogue stage, the output filter, and the on-board regulation. The output filter is very good and suits 4-6 ohm speakers so there’s no immediate need to change that. All I’ve done is to add Vishay MKP1837 10nF polypropylene caps in parallel to the output filter caps to increase their high frequency performance. This added a little more definition to sibilant sounds so they are more accurate and less slushy.

Here is the regulation circuit. The first regulator, LM5010, is switching and it’s so noisy they don’t give specs for it. What they do say is that output ripple is around 200mV for a 40V input. Yikes. The second reg is LM2940 which is the noisiest 3 pin reg available at 360uV, plus it doesn’t have the bandwidth to block the high frequency noise from the LM5010. I’ve used one of these in an amp before – not good – and it’s the reg that drives the gates in the amp so that is a great shame. The last reg, a 1117, is also noisy but it only drives external logic and LEDs, so it’s largely irrelevant to sound quality.


Here is the analogue circuit. It’s also disappointing in hi-fi terms since it uses a DC biased op amp (filtered 6V at the positive input) to simplify the power supply, but that means it needs DC blocking caps before and after the circuit, which degrades sound quality significantly, unless they are exceptional caps and these are not. The BOM in the EVM datasheet says they are Panasonic FK, 10uF aluminium caps, costing 40 cents each. The TPA amp also has DC on it’s inputs so it must use some form of blocking but 2 sets of caps is not the best. The two op amps are both inverting, unity gain, for single ended input so their function is only to convert SE 2v RMS to Balanced 4V RMS so they are part of gain structure and not critical. The amp itself is switching and generates noise so it would be good to isolate the amp inputs from other gear and avoid common mode noise issues. So, this circuit is not good, but it can be bypassed completely by using 10K:10K audio transformers. This is especially interesting part of this project since I am new to using small signal audio transformers – I found a good primer for this at Rod Elliott’s site.


I bought some transformers on Taobao  with these specs:

Frequency response: 20hz-20Khz
AC impedance: 10K: 10K (1 + 1: 1 + 1)
DC impedance: 170 + 170: 200 + 200
Inductance: 35H + 35H: 35H + 35H

I removed C17, C28, C55 and C63. I soldered a twisted pair of wires onto the pads that had been used for the negative cap terminals – these connect into the amp via a Mhz low pass filter with 100R in series. These wires went to the secondaries, and the primaries are the audio input: the simplest arrangement with no ground interconnect or centre tap as a first test with full isolation. This worked but added a tiny amount of background noise, so I connected the signal ground from the input primary to the centre tap of the output secondary and the noise disappeared. And the sound is MUCH improved, although it’s too detailed and requires some re-balancing of the tone. However, the definition is vastly better so if I can re-balance the signature, they are keepers, and it will be interesting to see if they can better Mundorf silver/oil cap coupling. I put the impedance and inductance values into a zobel calculator and it gave 0.45uF and 12.5kr. That has a corner frequency of 20Hz so it will alter the whole audible frequency range. So first I tried a 20k VAR with no cap and found that around 9kr sounded about right. I dug out my old scope to get this right, but found I could not accurately measure what I was hearing. So I kept the 9k for the time being.


The second EVM arrived and I hooked it up to start burning in. It was immediately obvious how much softer and duller the original amp is compared to the week-old one with input transformers. I’ll burn in this second one and use it for back-to-back listening tests against the modified one.p1080132After some burn in, I started modifying the EVM to get the best out of it but still using the op amp input circuit. I replaced the 5532 op amps with Ti’s top grade OPA1612, and replaced the Panasonic FK signal coupling caps with Wima MKS 4.7uF polystyrene caps, each with a Vishay MKP1837 10nF bypass in parallel. I also replaced the Panasonic FC 2200uF bulk power caps with Panasonic AMX  3300uF caps, again with Vishay bypasses in parallel. I added more bulk capacitance on the Sigma 11 regulator – increased to 10,000uF Nichicon KG after the rectifiers and 3,300uF Elna Silmic II on the output of the regulator. To ensure the output impedance stays very low, I added some thick cable directly from the power resistors on the Sigma 11 into the EVM. Taken together, the sound quality improvement is obvious and significant compared to the original EVM, so then the second amp was modified too.