Recently (actually several years ago, it's just taken me this long to write this up...) I bought a subwoofer and an all-in-one equalizer/crossover to add to my sound system. The only problem is that the equalizer has balanced inputs and outputs and my DIY power amp only has unbalanced inputs. I figured this wouldn't be a problem because I could just use TRS-to-RCA leads to connect the two, but as it turns out this causes a nasty ground loop (I'm not sure why, because I've never had any problems using the same leads between the power amp and mixer, but whatever...).
What I needed was a balanced receiver to convert balanced to unbalanced, kind of like a reverse DI. At first I though about using a passive DI backwards, but when I borrowed a DI to try it out, I found the XLR output connector was the wrong gender to be used as an input. While a gender changer would solve this problem very easily, I'd still need to buy a stereo DI to do it properly (or two mono DIs). More importantly, the transformer in a cheap passive DI would leave a lot to be desired in terms of sound quality, and I didn't want to spend a fortune on this, so instead I decided to build something to do the job.
There are a lot of ways to convert balanced audio to unbalanced. The best way is to use a transformer, but good transformers are expensive and cheap transformers sound shit, so instead I went with an active circuit. This can be done with a single opamp, but very closely matched resistors are needed otherwise the common-mode rejection will be terrible. Even 0.1% tolerance resistors are not really good enough. A better circuit (shown in the link above) uses three opamps, but its common-mode rejection ratio is still only 30dB or so.
On the other hand, a Texas Instruments INA134 has a (theoretical) CMRR of 90dB. From what I can tell from the datasheet it does this using the single opamp circuit, but uses laser trimmed resistors to acheive perfect balancing. In a real circuit I doubt its performance will be anything like that good, but it will still be better than anything you can make from opamps (and much better than discrete transistors). Of course a decent quality transformer would still be better (especially in terms of RF rejection), but INA134s don't cost £100. Also, I already had a pair of INA134s on my breadboard for a project I've been messing around with on-and-off for a while.
The INA134 circuit is taken directly from the datasheet. I decided that it would be a good idea to include gain controls, given that my power amp doesn't have them and there are no output level controls on the crossover. These are shown directly after the INA134s, followed by an opamp to buffer them from the load of the power amp. At this point I'm sure some of you are wondering why I'm using linear pots instead of logarithmic. The answer is that I'm using load resistors (15k) between the wiper and ground to fake a log curve that is actually better than most real log pots, which are really just two linear pots stuck together. You can see a good technical explanation of this here.
If I was doing this again, I'd add a coupling capacitor between the balanced receiver and the potentiometer. The crossover has a small DC bias on its outputs, which makes a crackling noise when turning the potentiometers. This doesn't matter too much given that this is a set-it-and-forget-it control, but I'd still be happier getting rid of it. The capacitor should be at least 1uF (ideally film) for a -1dB point of 20Hz. If you add this you probably don't need the output capacitor, because any half-decent opamp will have virtually no DC bias on its output, but you might want to keep it there anyway. If you do decide to keep the output capacitor, you might also want to add a 100k or so resistor between the output and ground to bleed off any DC bias and prevent popping noises when making connections. I don't think there will be any, and you shouldn't really make connections with the power turned on anyway, but add it if you want to. When a power amp is connected, its input impedance should bleed off the DC bias fairly quickly.
You might think the output capacitor is bigger than it needs to be, but remember that it will be in series with the input capacitor in most power amps, which reduces the overall capacitance. If the input capacitor in the amp is also 2.2uF, the overall value will be 1.1uF, giving a -1dB point of 10.3Hz. If the input capacitor is 1uF, you get 687.5nF and -1dB @ 16.7Hz. If you're connecting this to a preamp, the higher input impedance (47k is as close to a standard as you're gonna get) means you could get away with a smaller capacitor, but bare in mind the input capacitor on the preamp will probably be smaller too. In my case none of this stuff actually matters because I'm only using this for my main speakers and have a seperate sub, so I don't need response down to 20Hz. Even if you don't have a sub, your speakers probably don't get anywhere near 20Hz anyway (in fact, most subs don't even go that low).
The circuit will probably work fine with just about any opamp ever, so just use what you have on hand. I recommend an NE5532 as a happy medium between price and performance, and although I haven't tested it with one I have no doubt it will work just fine. The JRC4558 (or equivalent from other manufacturers) would be OK too. I even used an MC1458 for testing and it sounded alright. There's no need to go with high end opamps unless overkill makes you feel better and you don't mind spending money. That said, I'm using an OPA2132, mostly because I have a load of them lying around. I would usually recommend metal film resistors for minimum noise, but I built mine with standard carbon film resistors I had lying around and can't hear any more noise than normal. I don't think there's enough resistors in the circuit or enough gain for it to matter.
The power supply is designed for an external 15V AC transformer (wall wart), to keep mains noise out of the box. The input must be AC. The circuit can be changed to work with DC, but it would need a major redesign, and a fairly high DC input voltage and/or reduced headroom. If you want to do this you'll have to work it out yourself. Worst case DC current draw is about 100mA (from both rails, accounting for both quiescent and signal current, as well as the LED indicator), which means that the AC input voltage should be no less than 13.5V to prevent regulator dropout and no greater than 16.5V to allow the regulators to run without a heatsink. These numbers take into account capacitor ripple and the fact that mains voltage can vary by as much as ±10%. In reality, you'll only see that much current consumption with both inputs driven to clipping and both outputs shorted, in which case you probably don't care if the regulator drops out or goes into thermal limiting. Under normal use, DC current draw shouldn't exceed 25mA, meaning input voltages as low as 12V AC should be OK. The input should still be kept under 20V AC to avoid exceeding the 35V maximum for the regulators in case the mains voltage is a little higher than usual, especially considering that the regulation on small transformers can be as poor as 20% (meaning the secondary voltage could be as much as 20% higher than spec'd).
In my build I use a 15V AC transformer, though the secondary voltage measures over 18V no load and pretty much bang on 18V quiescent (my mains voltage was about 3—4% high when I made these measurements). This has worked reliably for several months now, with the regulators conformatably warm to the touch and absolutely no audible hum. (There actually is a slight hum, but that's because of an issue in the power amp which I still haven't gotten round to fixing). The input capacitors should ideally be rated at 35V (or 50V if the AC input is over 18V to allow a bit of safety margin). I'm actually using 25V capacitors because they're all I had in stock, but that's running them right at the edge of their ratings (I measured 24V DC), so I do plan to replace them someday...
In case your wondering if you can leave out the power LED... well, yes and no. When I first tested the power supply I couldn't get the negative rail to work at all—I measured around 20V, not the 12V I was expecting. At first I thought the 7912 was dead, but swapping it out for another one made no difference. I even tried disconnecting the protection diode thinking maybe it was shorted, but that didn't help either. I knew I'd come across this issue several times before, but I can never remember what the solution is. As it turns out, the 79xx series regulators need a minimum load current of 5mA or they won't regulate. The quiescent currents of the opamps and balanced receiver ICs may or may not make up this minimum, depending on the batch of chips they came from.
The LED is there simply to make sure the minimum load requirements are met at all times. If you don't want the LED, you'll still have to leave the series resistor in place, between -12V and ground. In this case you can increase the resistance to 2.2k to reduce the power consumption if you want (you can actually increase the resistance to 1.8k even with the LED, but it won't be very bright). The 78xx series regulators don't have this problem, so the load resistor doesn't need to connect to the positive rail. If you really want to keep the supply rails perfectly balanced, you can connect the LED and series resistor between the two rails, but you'll need to double the value of the series resistor. In theory this might reduce noise coupling from the supply rails to ground, but the difference will be so trivial it's not worth the extra power consumption. The 79xx also needs at least 25uF load capacitance on its output to remain stable (or 10uF if you go with a tantalum capacitor, which I don't recommend), but this isn't a problem because I always bypass the rails to ground with 100uF electrolytics anyway.
Ideally I would have built the circuit into the amp, either swapping the RCA inputs for TRS or XLR sockets, or leaving the existing sockets in place and adding a balanced/unbalanced switch. I decided against this because I needed the system to set up the music for a party that weekend, and I didn't want to find myself struggling to get the amp working come Friday night. At least if the balanced receiver wasn't finished in time I could make do with the ground loop, which wouldn't be audible over loud music anyway. So instead I built the unit into one of these enclosures. They have slots for mounting circuit boards, and the plastic end panels (and the aluminium) are easy to cut, although they do scratch a bit too easily. I would have liked to make a custom PCB, but I couldn't wait for the 2—3 week lead time, so I built it on veroboard instead. Once the parts showed up the whole thing was finished in a couple of days, and it probably could have been done in under a day if I wanted to put in the effort.
The circuit really is as simple as it looks. The power supply is to the left, the balanced receivers are in the centre and the output buffer is on the right. The twisted blue/white wires on the far left are the AC input, the red/black wires in the middle feed the power LED and the green/yellow wires are for the balanced inputs. The red/black and white/black wires on the right are the unbalanced outputs, and the twin shielded wires connect to the potentiometers on the front panel. I used shielded instead of twisted cable here because these cables are longer—the board will be at the back of the box, physically near to the connectors but away from the potentiometers.
When I'm working on veroboard, I prefer to use the non-copper-clad type and wire everything up point-to-point, because it makes routing easier. One disadvantage of this type of construction is that because you are really just soldering wires together, the joints can be quite big and use a lot of solder. This means you end up holding the iron on the joint for a lot longer to get the solder to take, running the risk of overheating parts. For this reason I don't recommend soldering DIL chips straight to the board like I would with a normal PCB, so I put them in sockets instead (the regulators I'm less worried about, because their TO-220 cases should dissipate the heat better).
With the board finished, it was time to move on to the enclosure. I started by placing the connectors on top of the back panel to make a mock-up, then when I was happy with the layout I marked the positions of everything in pencil. Next I drilled out the holes, starting with a small 3mm bit to make a pilot hole and gradually working my way up until the connector was easy enough to push in but still fitted tightly. This soft plastic is very easy to cut, but trying to drill a 10mm hole without making a pilot hole first is not recommended. As it turns out, I don't have a big enough drill bit to make the two biggest holes, so I had to drill them out with one of these things:
I'm not actually sure what this thing's called, but you can find them in a decent-sized drill bit set. I actually made this harder for myself than it should have been, because I didn't think it through until I'd already started drilling. It was only after I'd drilled out the hole with the biggest drill bit I could find that I realised it still wasn't big enough and I'd have to use one of these tools. Ideally, this tool needs a relatively small pilot hole to fit the pointy end into, then it should guide itself the rest of the way. Of course now I had a 10mm diameter hole to start with, which meant I had to guide the tool by hand. The end result is that it wobbles around as it cuts and you don't get a perfectly round hole, but you can't tell the difference once the connectors are in.
The next step was to clean up the holes with a file, rub the pencil markings off and then fit the connectors. From top-left to bottom-right, you can see the left unbalanced output (RCA), right unbalanced output (RCA), right balanced input (1/4" TRS), left balanced input (1/4" TRS) and AC power input (2.5mm barrel jack). You can also see that I haven't quite rubbed all the markings off, and there's a rather nasty scratch above one of the TRS sockets. This is the result of slipping while trying to deburr one of the holes, and ramming the file into the plastic. This soft plastic scratches way too easily.
Now it's time to do the same thing with the front panel. The two centre holes were also too big to make with a drill bit, but this time I remembered to only drill a pilot hole before moving on to the cutting tool.
Here it is with the power LED, left channel gain control and right channel gain control fitted. The LED is mounted in a bezel which clips into the hole, then the LED slides inside. I got these from Maplin years ago. The potentiometer shafts still need cutting down to fit the knob, but I want to test everything first.
Soldering it up was relatively quick. There's still one loose wire here. If you look just below the TRS connectors you'll see a blue wire which jumps from the shell of the AC input connector to the sleeve of each TRS connector, and then trails off into nowhere. This wire ends in a solder tag which will be bolted to the aluminium chassis when the whole thing gets put together.
Once I had the whole thing wired up, it was time to see if it actually worked. In this picture, I've got my phone (not visible) hooked up to a channel strip on my mixer, then from there I'm routing the signal out of an auxiliary send and to one of the balanced inputs. The unbalanced outputs are sent back to the 2-track input on the mixer, and after a detour through the master bus out to my headphones. This only lets me test one channel at a time (there's only one pre-fade auxiliary send on the mixer), but that's no big deal really. As a side note, I really don't recommend testing using headphones. If the circuit decides to oscillate, you could easily end up deaf. I figured it was OK here because I kept the headphone volume on the mixer fairly low, but in reality if things went pear-shaped it probably still would have ended badly.
Here's a shot of it nearly finished - just need to put the lid on. You can see how the veroboard fits into the slots extruded into either side of the case, which I guess must be a standard thickness (probably 1.6mm, given that's the thickness of a board you'd usually get from a commercial PCB house). There's a lot of empty space in their. I guess that's what I get for buying a box just because I've used them before, without putting any actual thought into how big the thing will be.
This is how I mounted the LED to the front panel. The plastic bezel has little clips on the back which are supposed to hold it in place, but I ended up drilling the hole ever so slightly too big and it didn't hold in very well. It wasn't so lose that it would pull straight out, but it span around in the hole very easily, which could end in the wires getting tangled and pulling the pins closer together until they short out. A bit of epoxy on the inside sorted that.