Matt’s Big Class A Amplifier (AudioXpress article) See the main amplifier page for the images and figures.
Summary:
This article describes the design and construction of a 20W per channel single ended class A audio amplifier with a bi-polar input and MOSFET output. It uses a three-stage topology with local and global negative feedback.
Introduction:
One channel of the amplifier was my senior project at San Diego State University. I presented the amplifier in December of 1999.The amplifier was inspired by my third and final analog design class. At that time, we designed and simulated an operational amplifier (op-amp) using Microsim PSPICE. We took all the transistor stages we had learned in previous classes to design a complete operational amplifier. The first two stages of the op-amp are almost identical to the first two stages of my audio amplifier. The changes to the input stage included the type of current sources (the ones in the simulation relied on regulated power supply rails) and a much higher transistor bias (current flow at idle).
The word op-amp sends some high fidelity fans running for cover. However, most op-amps are single ended (SE) class A amplifiers up to the output stage. Standard op-amps are biased very low to keep power consumption down to a minimum. With an increased bias and a class A output stage, an operational amplifier becomes a nice sounding audio amplifier.
I chose the output stage topology for two reasons. First, it is the most linear class of output stages. Second it more unique and interesting to design. A class A amplifier is made for do-it-your-selfers (DIYers). The heat sinks and the power supply capacitors make the cost of a SE amplifier high, but DIYers can often find parts at surplus stores.
Looking at the block diagram (fig 1 blockdiagram.gif), the amplifier is fairly straightforward. A differential input stage, common emitter gain stage, and an output stage make up the amplifier. The current sources in the full schematic (fig 2 fullschematic.pdf) can make the circuit more complicated looking than it really is. Let’s look at the Current source first.
I source:
The current source (fig 3 Isource.PDF) gives good current regulation independent of the variations in the voltages being supplied. Although explained before [1], let’s take a look at how the current source functions. Looking at the figure 3, at time 0 there is no current flow and therefore no voltage drop. The voltage on the base of Qx rises quickly and turns Qx on. Current then flows through Qx and through Rc. As current increases through Rc, so does the voltage drop across Rc. Eventually approximately .7V is dropped across Rc. That same voltage is applied base to emitter on Qy (ignoring the small Rd drop). This transistor turns on and starts to "steal" current away from Qx. This reaches an equilibrium where the current through Rc and therefore the transistor to be biased is .7V/Rc. The value of Rc is chosen to give the desired bias current.
Design:
The first two stages of the amplifier are straightforward. Please note that I consider the emitter followers (Q13,Q10) as part of the 1st and 2nd stages respectively.
The first stage is the differential input. It is made up of the PNP transistors Q1 and Q5. The input stage has a current mirror (also known as an active load) made up of transistors Q7, and Q8. The stage is biased by the current source Q4, and Q3. The 2nd stage is a common emitter gain stage (Q9). Mr. Douglas Self’s book had some suggestions for current biasing and high beta transistors [2]. Very similar values were used (differential = 4.3mA, emitter follower = 6.5mA). These two stages are very common and can be found in most operational amplifiers and audio amplifiers. They are described in great detail in many books. In short, the differential amplifier provides a convenient spot for feedback and provides some gain. The common emitter stage provides high Voltage gain. The higher the open loop gain (gain with out feedback), the more the amplifier gain will conform to the simplified op-amp gain equation (1+R2/R1) or in the case of this amplifier 1+R18/R15 (1+15KW /604W =25.83V/V).
The output stage:
I have read Mr. Nelson Pass’ articles regarding his SE amplifiers with great interest. The output stage in my amplifier is single ended, but it is very different from the stage used in the Zen amplifier. My stage is a source follower with a voltage gain of less than one. The Zen amplifier has voltage and current gain in its output stage.
The output stage specifics were determined by choosing the standard speaker impedance to be used. Knowing that the theoretical efficiency of a single ended class A amplifier is 25%, I chose to make an amplifier that was 20W per channel. That meant that with small additional losses I would need to dissipate about 100W per channel. The actual power draw was 112.5W. Since the output stage uses a constant current source, I needed to calculate how much current was needed. I intended to use an eight W speaker with the amp. I2 R=P therefore I2*8W =20W Since I wanted a constant 20W, the equation becomes (.707*I)2=20W…. I=2.24A. Using the constant current source discussed earlier .7V/Rx=2.3A=.313W . The nearest available resistor in a 5W package was .3W . This value caused 2.3 Amps to flow. I now had enough current to give 20W. I needed to be sure I had enough voltage. The voltages of the positive and negative rails are determined by V/R=I and the fact that the amp is designed for an 8W load. The equation becomes V/8W =2.24A. Solving the equation gives the minimum voltage of 18VDC. The amplifier can not swing to the rail because of base-to- emitter (Vbe) and gate-to-source (Vgs) voltage drops. The most important drop is the Vgs of the output MOSFETs. The gate to source threshold voltage is approximately 4V. Adding 4 volts to the required 18.0V, the new minimum rail voltage is approximately 22VDC. Accounting for the rail voltages drooping with a constant 200W load, I wanted to find a transformer that would give approximately plus and minus 25VDC rails.
The extra eight volts needed (four Volts for each MOSFET) translates into 18 wasted watts. A separate power supply with higher voltage could be used to make the output swing closer towards the plus and minus output rails, but that sounded like a lot of work for a small return. In an amplifier that burns over 110W per channel at idle, what is 16 more watts?
A speaker with a lower output impedance is going to cause less output power. Unlike Class AB designs, where more current is provided as necessary (given an adequate transformer) a class A amplifier only has so much current to give.Then the constant current source on the output runs out. Lowering the impedance of the speaker to 4ohms will half your power. Truly it will only half the power being delivered on the negative swing (where the output is less that 0V and all the current for the speaker must come from the current source).
Going back through the current and voltage requirements, you can see that it would be easy to modify the output power. I looked at many different MOSFETs before choosing the IRFP140N. The MOSFET has a 100V V drain-to-source rating and can handle 33A. They are readily available and reasonably priced. If more power is desired, you might want to consider using a fan. Even though the MOSFET can handle more current flow, you can only get so much heat away from a single TO-247 package with out forced convection. If you put several MOSFETs in parallel for increased power handling, you should use source resistors to help current sharing. Even hand matched MOSFETS don’t share well [3] above or below the current that they were matched at.
Board creation:
After the circuit had been simulated thoroughly, I used the Veribest PCB program (bought by Mentor Graphics) to lay out the board. The board was made using a machine that resembles a router. The single sided PCB was attached to a large metal plate and then the machine routed away the copper that was not needed. For example, if a singe trace is needed, the machine grinds away the copper to the left and right of the vertically running trace. The more it grinds away, the more space between that trace and another trace. The first board turned out the best. The second board was done about 6 months later. The amount of copper removed between traces was reduced to speed up the process. This resulted in many shorted traces. The ones I could locate I removed with an exacto knife. Two hairline shorts I could not find were removed with a 12V battery and a brief burst of current flow. That method is risky, but it was very effective. I had the machine drill all the holes the same size. On the first board, I enlarged the few holes that needed to be increased in size with a high-speed drill at school. On the second board, I used a standard drill with a bit that was slightly too large. I paid for this by having several of the copper traces lifted off the board. The board became ugly and messy, but still functions well. The board artwork is shown in figure 4 (ClassA_pcb.pdf).
Power supply:
I used a toroid transformer for its low magnetic radiation. A Toroid of Maryland #738.182 was used. It has a 385 volt-ampere rating and a dual 18AC output. The secondaries were put in series to achieve a 36VAC center tap transformer. The dual primaries were put in parallel for 120VAC use.
Originally I had two capacitors in parallel per voltage rail after the bridge rectifier. I used a pair of 29,000uF and 59,000uF 75V capacitors for a good reserve. I had bought them previously from a surplus store. 35V capacitors can be used. A slow blow fuse was used to accommodate the large current spike caused by the capacitors charging at turn on. I originally used a 2A fuse when I had just one channel completed. However, the fuse would sometimes blow especially if the amp was cycled on and off. I went to a 5A slow blow fuse. I now use a "pi"filter to reduce the ripple. See figure 5 (PowerSupply.pdf) for the updated power supply. The pi filter started out as.3 ohms of resistance connected between the first and second set of capacitors on both the plus and minus voltage rails. The bridge feeds the 29,000uF capacitors and then current flows through the resistors to the 59,000uF capacitors. Most of you have probably seen an inductor used in a pi filter, but with a Class A amplifier and its fairly constant current draw, a resistive pi filter works well [4]. I finalized the power supply by replacing the resistors with thermistors. The SG420 thermistors I used have a room temperature resistance of approximately 2W . When first turned on, the current rush is improved by making the 59,000uF capacitors charge through the thermistor. When the thermistor warms up, the resistance of the thermistors is approximately .15W . Because the thermistors have a 23A maximum rating and I only have 4.6A running through them, they never get to their normal operating temperature. This means their resistance never drops to the intended impedance. The use of a larger than needed thermistor gives surge protection and enough resistance to filter out a lot of 120Hz ripple. Two for one deals are rare. Looking at just the resistance of the thermistors and the 59,000uF capacitors, you can see a low pass filter with a –3db point of 1.8Hz has been created (1/(2*pi*.15W *59,000uF). The effect can be seen by the reduction in 120Hz ripple. Before the thermistor, the ripple on the positive rail is 760mVp-p. After the thermistor the ripple is 90mVp-p.
HUMMM:
By far the most frustrating problem was the hum the amplifier produced in the speakers. I first approached the hum problem with the theory that the hum must be coming from the 120Hz ripple (2*60Hz because of the bridge rectifier) on my voltage rails. I neglected to measure the frequency with an oscilloscope. I added more capacitance (as if the in-rush current wasn’t already big enough), and installed a "pi" filter. Neither helped. I did keep the pi filter as it did make a huge reduction in ripple. I finally noticed that the ripple on the output was 60Hz and not 120Hz. This showed me the noise was being introduced before the bridge rectifier. I unbolted the toroid transformer and rotated it while listening to the hum. I found the best spot for the transformer was on its side rotated off center. If I positioned the toroid just right, I could get rid of almost all the hum. However, if I moved the shielded input cable a small amount the noise would return. I resisted mounting the transformer externally, but I eventually did. Before I realized my hum was transformer related, I tried numerous changes to the board. I put another RC filter on the input and gain stage, moved grounds around, and tried changing capacitor values. I tore the underside of the board up a bit, but none of them helped since my problem was from the transformer. Photo 1 (amp_cover_on_PS.jpg) shows the amplifier with the small power supply box to the right. My hum is now very low. With no inputs connected, there is a 60Hz ripple of 3-4mV that is not audible with my 90dB/1W speakers. I believe that some of my hum problem was due to my RCA input jacks being mounted on opposite sides of the chassis. If the shielded input cables ran together, they would have very similar 60Hz noise on them and therefore little 60Hz potential between them. Since I ran my input cables far apart, the 60Hz noise was not the same on both cables. The problem was increased because the toroid is in between the left and right input cables. This could cause a small current to flow in the ground line and cause my hum. This is my theory and possibly the reason you always see RCA input jacks mounted right next to each other. My transformer is now in a plastic box with one 14 guage 3 conductor AC cord going from the secondary of the transformer in the plastic box to the bridge rectifier in the chassis. A second AC cord provides the primary with 120V. The second AC cord can be 18 guage since there is much less current with the 120V potential. The earth ground wire is not used because the transformer is in a plastic box. All voltages in the chassis are isolated from the primary via the transformer. The power switch and fuse on the metal chassis are no longer being used, but were left for aesthetic value. The functional power switch and fuse are mounted on the plastic power supply box
The hum problem is more of an issue on Class A amplifiers because of the large current draw at idle. Not only do the rails have more voltage ripple at idle, but also the magnetic fields produced by the transformer are stronger. With a class AB amplifier, very little current is drawn at idle.
The heatsinks have only the output and current source MOSFETs (M1 and M2) mounted to them. A sil-pad is used to thermally couple the MOSFET to the heatsink and electrically isolate the drain of the MOSFET from the heatsink and chassis. Photo 2 (board_in_amp.jpg) shows the stuffed board mounted in the chassis. The reflection of the coil, 5W resistor, and other componets on the board come from the shiny finish of the brass plate that holds two heatsinks together. I first tried a heatsink having approximately 1500cm2 of surface area. Even though the amplifier was not built, the heat dissipation was simulated by running 2.3A of current through a MOSFET that had 25V across it drain to source. Since I only had one quarter of the heatsink (one heatsink of four for a complete stereo amplifier), I tested with only one of the four MOSFETs. The test simulated the idle condition where the output of the amplifier would be zero and there would be 25 volts across each MOSFET with 2.3A flowing. After the increase in temperature had steadied, the heatsink measured 144 degrees F (62 degrees Celsius). The temperature was not acceptable. The MOSFETs have a maximum temperature rating of 170 degrees C, but the human factor has to be considered. A temperature of 144 degrees F is painful to the touch. Temperatures between 130-140 degrees F were found to be the changing point between hot and painfully hot. A lower temperature was needed. Another heatsink was purchased and tested. Its surface area was 2300cm2. The heatsink temperature under test was 106.1 degrees F. This was appropriate for human contact so I bought three more heatsinks from the surplus store. The finished amplifier causes the heatsink temperature to be between 120-130 degrees F at normal room temperatures. The heatsink is very hot to the touch, but it will not burn you and can be touched with slight discomfort. I would recommend even more heatsink surface area. The cooler the FETs run the longer the life they should have. Four heat sinks were used (two on each side). I connected the pairs of heatsinks together with a brass plate.
Chassis:
The heatsinks were used as the frame of the amplifier. Photo 3 (amp_back_top_off.jpg) shows the opened amplifier in its final configuration. All aluminum panels bolt to the heatsinks. I used .125" aluminum so the walls would not bend with the weight of the transformer and large capacitors. The transformer was later removed from the chassis. This was my first time working with metal. I was learning as I went. I do not give dimensions for the metal, because the heatsinks were bought surplus and it is unlikely the exact ones could be found again. The chassis was brushed and then chem-coated a gold/yellow color. I used five way binding posts for the speaker outputs.
The boards are mounted along the left and right long sides of the chassis with two stand off posts. The other side of the board is supported by the MOSFETs, which are bolted to the heatsinks. Mounting the output transistors directly to the heatsink allowed less wiring and kept the signal path of the gate drive short. I punched vent holes in the top of the amplifier. I should have used thinner metal for the cover to allow for easier punching. I bent and marred my cover a little when I was hanging from the punch lever trying to get through the .125" metal.
Bringing the amplifier up:
Simulations can only take you so far, but as evidence that a simulation can get you started in the right direction, I am happy to say the amplifier worked the very first time I tried it.
A large value resistor should be substituted for R28 (3.3-4.7W ). This will keep current flow low in the output stage and therefore power dissipation low. Fewer electrons will be available for cooking a mistake. Hook up the PS to the AC line (slowly brought up on a variac would be really nice). I usually use 8ohm resistors on the plus and minus voltage rails before they enter the board. The 8W resistance is enough to keep a short on the board from causing mass component destruction, but low enough to power the circuit (that is if you are using a high value for R28).
Feedback: I consider the amplifier to be DC coupled, but there is room for argument. The frequency response starts to roll off at 20Hz (-3db at 1.7Hz). The amplifier does not respond to DC because of C1 and C10 in the feedback network. These capacitors block extremely low frequencies from being attenuated by the divider made up of R18 and R15. The full DC offset is fed back to the differential section. This helps give a low DC output offset. Higher frequencies are passed through the capacitors to ground so only a percentage of the output signal gets sent to back to the differential input. The amplifier is DC coupled input to output, but the feedback network keeps the amplifier from having true DC response.
Measurements/data:
The input impedance or frequency response is not as important as the sound of the amplifier, but is nice to have some quantitative numbers. Here is how I measured a few common amplifier specifications.
The gain of the amplifier was measured by giving the amplifier an input that caused the output to be 6VRMS. The input was then measured and found to be .227VRMS. (.227VRMS)*(Av)=6V RMS. The gain is 26.43 volts per volt (V/V) (only .6V/V off from the calculated gain using R18 and R15 values).
The bandwidth of the amplifier was measured by increasing the input until the output was 4VRMS at 1KHz. The –3db point is (1/( ))*(Vrms) = .707*4VRMS=2.828VRMS. The input was swept up in frequency until the output was at 2.828VRMS. The frequency was 201KHz. The input frequency was then lowered until 2.828VRMS was reached. The frequency that occurred at was 1.7Hz. The output voltage offset was measured with the RCA inputs left open. The left channel output was -16.5mV. The right channel had 1.1mV on the output. The Zobel network (also called Boucherot cells) is made of R29 and C9. The 10W resistor and.1uF capacitor (R29, C9) presents the amplifier with a high frequency load. L1 is the output coil. It helps protect the output from capacitive speaker loads. The output coil forms a low pass filter with the speaker. The inductance of the output coil in this amplifier is approximately 2uH and was hand wound. I used 16-gauge magnet wire with 23 turns. The diameter of the coil is 1.5cm and the length is 4.2cm. The exact value is not critical. I wrapped the wire around a large ink marker to get the shape. The coil should not have a core (only air).
The distortion graph shown in fig. 6 (dist_20W.GIF) was taken before the second channel was created. This means that a little more voltage was available and much more 60Hz and 120 Hz noise was present compared to the amplifier as it is now. The distortion is less than .02% THD at most frequencies and increases to less than .04% at 7.5KHz. I do not know why there is a bump in the distortion at 7.5KHz, but I would like to know if readers have any suggestions. The pi filter and external mounting of the transformer helped with the noise.
Grounding:
I followed a star grounding scheme. From photo 4 (grounding.jpg), you can see four black Phillips head bolts with a single light colored bolt in the center of them. The center screw is the center tap of the transformer. The large capacitor charging current spikes travel in this section. The grounds for the speaker outputs and on-PCB filter capacitors form the "star" ground configuration a few inches below on the plate. The two wires below the star are the grounds that go to the signal input (the shield of the input cables). Although aluminum is not the best conductor, the large plate (left over from the chassis) has very low resistance.
My input jacks are mounted to the back of the chassis and are connected to ground via the shielded cable. I tried lifting the connectors off the chassis, but I did not get improvement in output noise. As mentioned before in this article, the chassis does not have an earth ground connection.
Parts:
The parts placement is shown in figure 7 (class_A_parts3.pdf). The spots marked TP1 through TP6 correspond to signal ground, speaker output, filter capacitor ground, –25V, +25V, and input signal respectively. The on board filter capacitor grounds are separate from signal ground to keep the capacitor charging currents off the ground traces of the board. The two lines on near TP6 represent two jumpers. The jumpers can be leads trimmed off resistors or capacitors. The parts list is shown in figure 8 (class A values_MT.xls). All parts are from Mouser Electronics (http://www.mouser.com) except for the MOSFETs which are from Digi-Key (http://www.digikey.com), transformer (Toroid of Maryland http://www.toroid.com/), and the fuse holder and AC switch (Radio Shack http://www.radioshackcorporation.com/). All resistors are ¼ W except where noted. The Mouser part number for the ¼ W resistors are 271-VALUE where VALUE is the value of the resistor (271-1.5K is a 1.5KW ¼ W resistor). I listed a Mouser part number for a Keystone-Thermometrics thermistor that is similar to the RTI SG420 that I used. The Thermometrics part is rated lower in current, but it still should have enough resistance when warm to give a noticeable drop in 120Hz ripple. If you would like to use the RTI part, try http://www.pcipci.com/. The capacitors I list for the power supply should give similar results to the surplus capacitors I used. The chassis design and heatsinks are up to you. Figure 9 (transistor_pin_out.GIF) shows the pin out for both bipolar transistors used. Please note that the Mouser web site had a specification sheet for the 2SA970 that did not match the 2SA970 that I was sent. Figure 9 shows the correct pin configuration for the transistors I received. The MOSFETS have a standard pin configuration. With the metal back away from you and the legs down, the pins are Gate, Drain, and Source from left to right. 16 guage wire was used in the amplifier except for the AC cords.
For my description of the sound of the amplifier, I simply say it sounds good.
1. Nelson Pass, "The Pass Zen Amplifier" The Audio Amateur, 2/94 pg. 14
2. Douglas Self, "Audio Power Amplifier Design Handbook",1996, Pg. 75
3. Motorola TMOS power MOSFET transistor data, Q4 1992, pg. 2-7-15
4. Shrader, "Electronic Communication", Fifth Edition, pg. 179