Parallel Amplifiers
Instead of using the two halves of a class-B amplifier in the feedforward circuit there is no reason why we could not use two complete separate amplifiers. This is the traditional feedforward method, but this is generally more complicated, so why bother? The whole point of the improved class-B circuit is that no additional error amplifier is needed.
The advantages of an integrated circuit amplifier have been mentioned before, and in this case the use of two amplifiers such as the TDA2050 is not a great complication or expense. A single amplifier has a limited output current, insufficient for bridge amplifier use without some sort of current boost circuit as suggested in the bridge amplifier design note. The use of two TDA2050s in parallel could give 10A peak current with no additional boost needed. Of course it is not a good idea to just connect two i.c.s directly together. They would almost certainly not have well matched gains, and a high current could flow between the outputs. Using a pair of output resistors, and seperate feedback networks, however, parallel operation may be possible.
To obtain the advantages of feedforward error correction the usual method is to operate one amplifier as a small signal error amplifier, and make sure the other amplifier can provide virtually all the output power. The circuit I present here uses a slightly different approach. Both amplifiers can give high power output, but at small or medium power output only one provides the power while the other has only a small error correcting output current, and is therefore hopefully more linear. When more power is needed than one amplifier alone can provide, the error amplifier will provide extra current and will double the current available. A simple circuit is shown here which has not yet been tried.
The amplifier can operate from a plus and minus 30V supply using the LM4766 dual power amp i.c. shown here to give 40watts into 8 ohms. The OPA604 needs a lower supply voltage, plus and minus 15V appears to be the optimum value. A simple zener stabiliser should be sufficient. For clarity the supply connections and smoothing capacitors are not shown. The LM4766 has a mute function, and this can be used or disabled as required. Links to the date sheets are given at the end of this page.
The circuit is more or less identical to the improved class-B circuit apart from the amplifiers being complete i.c. power amps rather than the two halves of an output stage. The lower amp is the error correction amp, and the top amp provides the whole output power unless more than 4A is needed by the load. At low power levels the output current from the lower amp is very small, just the error correction current and any error from inexact component balance
The advantages of this circuit are the lack of any adjustment, the inclusion of overload and thermal protection within the i.c. amps, the relative simplicity, and the low distortion of a feedforward circuit.
The power amp i.c.s alone have a relatively poor offset characteristic, and the outputs could be at sufficiently different voltage to enable excessive current to flow through the 0.4 ohm output resistors from one amplifier to the other. The OPA604s are included partly to reduce this problem, and also to improve the noise performance a little. The max offset of 3mV is low enough, and being a fet op-amp the input current is very low and adds little extra offset. Ultra low offset op-amps such as the OP-27A/E (max offset 25 microvolts) could be used, but these have a higher input current which more than cancels the advantage. To reduce the component count further the dual version of the op-amp, the OPA2604 could be used. The amplifier then consists of just two i.c.s and a few passive components.
Apart from offset problems there is another reason for excessive current flow between the power-amps, and this is the variation in resistor values. Even though 1% tolerance metal film resistors are cheap and readily available, this is still not accurate enough to ensure sufficient balance to reduce the error amplifier output to insignificant levels. One option is to use 0.1% tolerance, which is expensive, another is to include an adjustment, which cancels a major advantage of the design. Another possibility is to add a fixed resistor, shown as Rx, which increases the gain of the error amplifier a little and can ensure that in the worst case the error amplifier output is still in phase with the other amplifier, and so it drives a little extra current into the load rather than causing a current flow between the amplifiers. The value of Rx could be calculated by assuming all resistors are 1% in error in just the right direction to give an out of phase error voltage, and finding the value of Rx needed to exactly cancel this. In practice such a large error is highly unlikely, and anyway a small out of phase error would not be a major disaster. Around 20k would be my own guess for a suitable value.
The circuit has not yet been built and tested, and I have not done any precise calculations of stability margins etc., so it is still nothing more than a suggestion. The simplicity and lack of adjustments make it an attractive design. The LM4766 is already a good amplifier, but this circuit could make it into a really excellent amplifier, with lower distortion, lower noise, and much better at handling difficult loads.
Click here for a link to the OPA604 data.
Click here for a link to the LM4766 datasheet.
