A HiFi Power Amp

A HiFi Power Amp

The recent aquisition of a pair of vintage Infinity RS-5b speakers prompted a search for an amplifier to drive them. According to the documentation that came with my speakers, an amplifier between 35W and 135W is recommended (not my 10W at 10% THD piece of garbage Sharp 3-in-1). Initially, I looked at commercial amplifiers (Yamaha, NAD, and Rotel), but was disappointed at their fairly pedestrian distortion figures. Thus a new hobby project was born :)

• Stereo. Whilst I'll use it for watching movies, my flat just doesn't have the room for a full set of surround sound speakers. My main motivation is listening to stereo music sources, so a stereo amplifier is appropriate.

• Low THD and IMD. In audio-speak, this translates either as clinical or as accurate. I'm an engineer by trade, so prefer terms like THD over "warm" or "cold" which could mean anything. Given that most of the commercial amps offer 0.02% THD or thereabouts, I figured a good target was 0.001%. This means the harmonics and intermods will be 100 dB below the fundamental, and will thus mean that the system performance will be dictated by the source (CD) material, rather than the amplifier.

• Ample power. 100W seems a reasonable amount. I currently live in a two bedroom flat, and don't want my neighbours to kill me :) However seriously, I figured 100W gives me a clear 20dB headroom at a nominal listening level of 1W.

• Low noise. I usually listen to my music at reasonably low level, so it's important for the amplifier input to have a low noise contribution.

• Moderate cost. I'm happy to blow around $1000, as long as I get plenty of enjoyable hobby hours (and listening hours) as payback.

• Good looks. This thing will (along with my preamp) live in my loungeroom. That means it needs to fit in. I don't want something that looks like an escapee from my shed, so a significant part of the design is involved with building a nice case.

• Useability. This pertains more to the preamp, but I wanted the whole thing to be completely remote controllable.

Design Background When I was a kid, one of my favorite monthly reads was ETI magazine. In January 1981, they published a series of articles describing the ETI477 MOSFET power amplifier, designed by David Tilbrook. This monoblock formed the basis of the "series 5000" HiFi amp. I desperately wanted to build one, but being all of 9 at the time, it wasn't going to happen. Optimisation There were a couple of drawbacks with the design. Firstly, it was based on obsolete TO-3 packaged lateral MOSFETs, and secondly the PCBs (like many kitset boards) were a pretty poor design anyway. I set to work redesigning the PCBs around modern flatpack equivalents (Hitachi 2SK1058 and 2SJ162). While I was at it, I swapped many of the remaining transistors for modern (faster) equivalents.

• JFET input diffamp: SST404 (SO-8). Was ECG461.

• Low power bipolars: MMBTA06/56 (SOT-23). Were BC547/557 and BC639/640.

• Medium power bipolars: MJE340/350 (TO-126). Unchanged.

• Power MOSFET drivers: 2SK1058/2SJ162 (TO-3P). Were 2SK176/2SJ56 (AEM6000) or 2SK134/2SJ49 (AEM6005 and ETI5000).

In order to dissipate 100W in an 8 Ohm speaker, one needs to put 28V RMS across the load. That's 40V peak. At the peak (assuming a resistive load) the amplifier needs supply 5A. Doing the SOA sums (more later) means that 2 pairs of drivers are needed. Further, the V_gs for the MOSFET can be around 10V at high current. This means the supply must be at least 10V greater than the peak output voltage. A twin 40V transformer is appropriate, with a peak secondary voltage of +/-56V.

• Use the transistor over a very small operating range.

• Use feedforward to cancel distortion (symmetry).

• Use feedback to cancel distortion.

Pretty much all amplifiers use a combination of the three. Feedback has a bad name amongst "audiophile" types. Poorly thought out feedback (especially across multiple stages) can result in oscillation (usually at very high frequency, which isn't audible in itself, but destroys the performance of the amplifier. Feedback needn't all be global though. A robust scheme involves linearising each stage of an amplifier independently (for example with emitter degeneration), then using overall feedback (with appropriate compensation) to set the gain.

I came up with some nominal specifications:

• The neat thing about the series 5000 is that it was built around new (at the time) Hitachi lateral power MOSFETs. Most power MOSFETs (VMOS, trenchFETs, HexFETs etc) use a vertical structure, where the current flows vertically. This has the advantage of stunningly low R_ds and hence high efficiency, but does nothing for linearity or capacitance. Lateral MOSFETs are a much simpler structure, where the gate oxide is formed on a flat substrate, and the current flows across the substrate. This results in well defined, controllable device parameters, good linearity, and relatively low gate capacitance. However, the R_ds of lateral MOSFETs is nothing to write home about. Most amplifiers at the time (and now as well) used bipolar output drivers. Bipolar transistors are cheap and plentiful. They have relatively high transconductance, and can operate reasonably fast. However they have some drawbacks when used at high power. The main one is thermal runaway. The gain of a bipolar transistor increases as it gets hotter. That means that if there's any imbalance between output transistors, the hottest one will pass most of the current, getting hotter until it ultimately expires. MOSFETs don't have this problem. Their gain decreases with temperature, so they share the load well. MOSFETs also have a high input impedance at low frequencies, and are capable (when driven by a suitable source) of extremely high slew-rates. Of course this very attribute makes them rather prone to HF oscillation when improperly compensated, but with careful design they're capable of impressive intermod performance. So having decided that now was the time to build a MOSFET amp, I wandered into the library at work and dug out the old ETI series 5000 amp articles, and had a read. I subsequently found that the series 5000 wasn't Tilbrook's final word on MOSFET amps. In 1987 he revisited the topic for a new magazine, Australian Electronics Monthly. This time (with the AEM6000 amplifier) he went all-out, with a matched-JFET differential input stage, and a complementary symmetrical voltage amplifier stage. A quest for super low distortion figures, with heaps of available gain. This looks like a good place to start.

I made the following active device substitutions:

• Now that I'd changed the transistors, I had to play with the values of most of the other components as well, in order to get reasonable performance while maintaining stability. Firstly, I decided that rather than the usual 1V RMS full-power input, I'd increase this a bit, to 4V RMS. This allows me to use more of the available dynamic range of my preamp, and requires a gain of 7.2, or 17dB.Transistors (and valves) are inherently non-linear devices. They must be linearised, or else they'll distort the sound. There are three ways to achieve this goal:

• I used Linear-Tech's free spice simulator to redesign the circuit around the newer parts. My main changes were to increase the emitter degeneration in each stage, to improve the linearity of each stage, at the expense of available overall open-loop gain. This is an approach that makes for an easily stabilised amplifier.A somewhat simplified schematic is shown below. Yes, it's a wonderfully complex beast of an amplifier. Heaps of symmetry, and plenty of stages, for ample open-loop gain. The schematic doesn't show the AEM6000 amp. It's my take on Tilbrook's design. The topology is the same, but the component values are different. For schematics of the AEM6000, you'll have to visit the library. Click on the schematic shown for a .pdf version of the real thing, including power supply decoupling and gate protection zeners etc.:

Please note that the schematic shown is that of the physical amplifier, after some tweaking on the bench to improve stability. For reference, the amplifier developed through simulation used 220 Ohm MOSFET gate resistors, a 6.2K feedback resistor, and 15p miller compensation capacitors on the voltage amplifier. The results of optimisation are fairly pleasing. Firstly, with the input filter and feedback resistor removed, there's plenty of open-loop gain (>90dB). Even at 10KHz, we still have 60dB gain. If we set our gain to 7.2 (17dB) the phase margin is around 70 degrees.

When the feedback loop is closed, the gain curve is pleasantly flat out to 2MHz or so, with very little peaking.

Adding a 1nF cap to the input defines the -3dB point at 160KHz. The gain is flat within 0.1dB to 24KHz.

Using +/-56V supplies, A full-power (4V RMS input) 1KHz sine-wave comes through nice and cleanly, with 0.0010% THD.

A full-power (4V RMS) 10KHz sine-wave is still clean, with 0.0011% THD. This promising result indicates that the amp isn't suffering from slew-rate limiting.

The intermod performance is also fairly good. Feeding a full power 10 KHz and 11 KHz sinewave, the IMD products are 100dB down:

The simulated amp behaves nicely under overload. Here's the output with an 8V RMS 1KHz sinewave input:

PCB Layout

In order to be faithful to the simulations, I took great care with the physical layout of the amplifier. The PCB is double sided, 6.2" by 3.6", and is designed for manufacture using most through-hole plated processes. Minimum track and space is 15 thou. I've used 4oz plated boards to minimise wiring resistance. It uses mainly SMD (low lead inductance) parts, with the exception of the higher power resistors, capacitors, and transistors. I used metal film mini-melf resistors, as I've had good results with these at RF frequencies, and poly and mica capacitors to ensure minimal distortion creeps in through capacitor nonlinearities.

Thermal Design

The thermal design also required some attention. A 100W class A-B amplifier needs to dissipate significant power. It'd be a terrible pity to design an electrically good amp, only to have it blow up the first time it's cranked up. The power supplies are chosen to deliver 100 W into an 8 Ohm speaker. However, considerably more power can be delivered into a 4 Ohm speaker. Using LTspice, I was able to calculate the power dissipated in the output transistors simply by measuring their I_d and V_ds. For this amplifier, with +/-56V supplies, worst case dissipation occurs at 150 W into 4 Ohms, of 155 W. This is shared equally across the four output transistors.
The transistor's maximum temperature is 150 degrees C. The transistors I used have a 1.25 K/W thermal resistance between the die and surface of the case. Shared across all transistors, this equates to a die-to-heatsink thermal resistance of 0.3125 K/W. Assuming a perfect heatsink, we could dissipate 400 W. The maximum heatsink thermal resistance to dissipate 155 W is given by:

I chose a 75mm x 300mm x 48mm Conrad cast heatsink, with R_th of 0.37 K/W for each channel. These will form the sides of the case, and result in a relatively compact amplifier. The thermal resistance is adequate (just!) when used with 4 Ohm speakers, but is plenty low enough for use with 8 Ohm speakers. Due to the restricted height, I've mounted the four MOSFETS to a 40mm L shaped aluminium extrusion, which is then bolted to the heatsink.

The PCB is designed with all the power transistors mounted underneath. Bolts pass through the PCB, then the transistors, then into a heatsink, as shown in the diagram:

Tweaking

After building the amplifier, some tweaking was necessary to ensure stability. Clearly the models aren't perfect. Occasionally power-up transients would cause the amplifier to oscillate at 1.1 MHz. I found that adding an extra 15p (to 30p) of Miller capacitance on the voltage amp stage greatly improved stability. At the suggestion of some of the experts on the DIYAudio forum, I also increased the MOSFET gate resistors to 680 Ohm and 470 Ohm for the N and P channel devices respectively. I also reduced the feedback a little, by substituting a 15K resistor (24dB gain) rather than the 6.2K one.

The results of tweaking are an amplifier that's stable driving an 8 Ohm load with a 2.2uF MKT capacitor across it, clipping hard (with the capacitor making some cool noises).

Measuring the Beast

Measuring the amplifier's distortion levels were (and are still) challenging. I borrowed a Tektronix SG505 oscillator, and AA501A distortion analyser. This combination is able to measure some pretty low levels.

First, here are plots of THD+N at 1 KHz and 100 KHz, while varying the power from 10 mW to 140 W (into 8 Ohms). The increase below 1 W or so is the noise floor of the distortion analyser.

Maximum power is 120W (0.0015% at 1KHz).

Repeating the exercise at 100 W into 8 Ohms, and varying the frequency gives us the following plot.

THD is about 0.0013% at 1KHz, and rises to 0.0048% at 10KHz. The increase below 100Hz is due to a high pass in the distortion analyser increasing its noise floor. I didn't quite hit my target of 0.001%, but I'm happy with the performance nonetheless.
With no input, I measure 52uV RMS on the output (80KHz bandwidth). That's 12nV/sqrtHz voltage noise at the input. Given 120W maximum power (31V RMS into 8 Ohms), that implies 115dB SNR.

In the flesh, the monoblocks look reasonably impressive. The heatsinks are certainly hefty:

Here's a photo of the assembled amp, with rear panel and bottom made from steel sheet. Unlike a lot of stereo gear, there's not a lot of empty space in the box. It's surprisingly heavy, too.

Listening

Now for the fun part. After setting the bias and offset trimpots with a dummy load (and testing the THD specs), I hooked up my speakers, connected my (as yet unfinished) preamp and a CD player, and stuck a CD in. I'm not really into audiophile adjectives, but suffice it to say I'm pretty damned pleased. It sounds every bit as transparent and clear as I hoped. It's stunningly quiet. I can't hear so much as a glimmer between tracks. I was pleasantly surprised to hear really low frequency detail, especially in movies. Explosions rock the room with wonderful rumbles.
One of the unanticipated gotchas is that I tend towards listening at rather higher levels than I should. Unlike my old amp, which made it quite clear when the volume was excessive, this one just keeps on trucking. No distortion, no mud, just clear (loud) music.
One small problem. When I turn the power off, I get a really impressive thump as the power supplies collapse. I've got some space in the box to add a pair of relays, and a circuit to disconnect my speakers when the power is shut off.
Here's some design files:

• Schematic in LTspice format.
• Schematic in pdf format.
• Schematic in Protel format.
• Board layout in gif format.
• Board layout in Protel format.

A quick word to the reader. I'm presenting this design as a blog. It's something I've done for myself, and I thought it would be nice to share. I'm not interested in selling kits, or boards for the project, so don't bother asking me. Further, this is a very complex amplifier. It's got lots of gain, many stages, and a huge number of components. It's not suited to building on veroboard, tagboard, breadboard, or any other crappy DC prototyping-on-the-cheap construction method. It has gain out to HF frequencies, so requires construction methods that match. That means a real PCB. So if you email me with questions about where to get a kit, how to build it with sticky tape and rubber bands, or other inane topics, please don't be offended when you don't get a reply.