What I did with TPA6120

The forum is called Headphone Amps (full-size), this particular amp is hardly very big, but it runs off the mains, so it doesn't belong in portable.

About a week ago a couple of TPA 6120A2 chips arrived from TI. I immediately set about building an amp with them, this is what I did.

It's not a very sophisticated approach, but I wanted a baseline design that I had a high level of confidence would work, because, naturally enough, I wanted to hear the chip. There are a lot of warnings in the datasheet about the bandwidth of the amp and it's potential for oscillation if certain layout considerations are not observed.

This is what the layout looked like:

It's a mix of SMT and thru hole components, but then it certainly would be anyway, since it includes the TPA6120 and a transformer. The actual choice of components was strongly influenced by what I had on hand and what I could get from the local Maplin's on short notice. What I could get included the 7812 and 7912 leaded regulators and the polyester input caps, what I had was LME49710 opamps, 1206 5% resistors and the 47uF SMT tants.

This is what it looks like:-

It needs built into a box with a volume control and a fuse. My impressions are that it works extremely well. I have optimised the PCB layout for home construction, with good clearances to the ground planes, and removed only such copper is necessary in order to economise on etchant. It's not an easy build, but it's entirely possible. Apart from the SMT issues, I had to drill a 3mm hole in the PCB below the centre of the chip that I could puddle solder in to connect to the heatsink pad on the bottom of the chip. Best not to hang around when doing this, but get it done promptly.

Subsequent to this working build I have revised the circuit with both internal and external feedback loops. If anybody wants to mess around with it, I have posted the new circuit on my website in the 'techno' section http://wakibaki.com/techno.php, it's at the bottom of the page below the reclocking DAC. The circuit can be built as per the example above, which I know works, or with the feedback loops implemented as desired, zipped gerber files are available for download. Undertake the build at your own risk, although I don't think the components totalled more than UKP25 (25 pounds UK). The gain-bandwidth product of the 49710 is 55MHz, so even with the gain of ~50 as I have redrawn the circuit it still has gain out to over a megahertz  If anybody wants the less sophisticated but proved layout, then I'll post the gerbers for that.

When I get around to building the new version I'll post a verification of the layout.

w

It sounds pretty much beyond reproach. Effortless. As good as or better than anything else I've ever heard. I'm listening to Chris Isaak's Baja Sessions which is a largely acoustic album full of tone at the moment, it's just exquisite. The phones are Shure E2C in-ear, which are known for having a very flat response and flat input impedance. I was listening to Van Morrison's Astral Weeks previously, a difficult album with which I am very familiar after 43 years. You can hear the presence of a track just disappear into silence when it ends. The only problem with that album is the difficulty in paying attention to the medium, the message is so strong.

I don't hear a lot of differences between amplifiers, to be honest. I'm not deaf though, and I've been a musician for a long time. Obviously there are differences between instrument amps, but where hi-fi is concerned I just like to settle down and enjoy the music.

I built this amp because I have the skills, having worked as a design engineer, and the specifications of both the TPA and the 49710s are extremely good. It's nice to think that the amplifier you're listening through is extremely accurate, but more because I can ignore it now.

I probably haven't extracted the absolute limit in performance yet but that's an unknown without instrumented test, I doubt there's an audible improvement to be had. I don't know if I can be bothered to do the testing, it might already perform better than I could measure with what I've got here.

w