Current sources are basic building blocks of almost every analog audio amplifier. There are countless circuits and each has its own unique properties. In this article I will compare a small selection of different variants with each other in order to select one for my next audio amplifier. Criteria considered in the comparison are: Thermal drift, supply voltage dependence, PSRR and step response.
In general, both source and sink were investigated. For simplicity, only the schematic of the source is shown. Plots may show results for the sink in case it deviates from the source.
Very simple current source using a LED as reference. Variant 1A uses a Zener diode instead of a LED. I guess this are the most common current source seen in many amplifiers. The LED mostly compensates the transistors temperature coefficient, resulting in low and flat drift.
Transistor reference current source with amplified negative feedback. Seen less often in amplifiers. This type is sometimes believed to have better performance. It is prone to instability and therefore requires measured to keep it stable. Strong negative temperature coefficient.
Since I found both circuit 1 and circuit 2 temperature coefficient having too much effect on the whole amplifiers bias, I investigated circuit 3. This is an attempt to improve thermal stability by using a LED plus a diode connected transistor as reference. The effect is that the coefficient is not lower than circuit 1, but it is reversed, i.e. becomes negative with slope roughly like circuit 1.
This circuit is made up of two current sources actually. The idea is that each source provides constant current for the reference of the other source in order to make the design independent from the power supply voltage.
Cascoded variant of circuit 4. Circuits 1 to 3 could be cascoded as well and this will dramatically improve performance. However, I concentrated more on the supply voltage independent circuit 4 for further improvement. The cascode transistors have base series resistors as an attempt to prevent the cascodes from oscillation.
Same as circuit 5 but without series resistors at the base of the cascode transistors. Instead, the cascodes feature emitter resistors. Also, the R-C series element between input and output and the joined base nodes aim to prevent oscillation.
Same as circuit 6 but without the diode connected resistor in the reference path.
All current sources were set to 1mA, 5mA and 10mA at 50°C for comparison. For curcuit 1 to circuit 3, the current of the reference was fixed at roughly 10mA for all scenarios. This temperature drift simulation is merely a crude approximation because in reality, the components are unlikely to be at the same temperature. However, it gives a hint how the circuits may behave.
Circuit 1 shows a slight positive coefficient. Replacing the LED with a 4.7V Zener diode (circuit 1A) leads to a temperature coefficient with less variation between source and sink, but more drift.
Circuit 2 has a strong negative coefficient.
Circuits 3 to 6 use the same method for compensation and show very similar coefficient. Plots 4 and 6 overlap each other, 5 is not shown as it is redundant. Circuit 7 shows a more flat and also negative coefficient.
An interesting detail is that the coefficient for source and sink of circuits 1 and 2 are different, while for circuit 3 to 5 they are identical.
One major drawback of circuits 1 to 3 is that their reference voltage and therefore, output current, is dependent on the supply voltage. The only solution for this seems to also apply a constant current to the reference. Circuit 4 does exactly this by mirroring the output current into the reference. I set both current paths to the same value. The effect doing so is dramatic: Circuit 4 delivers a near perfectly constant current from 5V up and circuit 5 from slightly below 10V up due to voltage lost in the cascode references. The other circuits show severe dependency of the output current on the supply voltage.
Power supply rejection is more or less acceptable for all circuits, but shows a trade-off for some circuits. This is related to the supply voltage dependency of the output current. Some circuits lack good PSRR at high frequency, while other perform better.
Step response shows dramatically different behavior of the different circuits. Circuit 1 and 3 perform best due to not using any amplified feedback. Circuit 2 shows severe ringing, but could be stabilized either with a base series resistor for the reference transistor or capacitors between some nodes. Circuit 4 and 5 show overshoot but no ringing. Base stopper resistors for the cascodes in circuit 5 seem to make the response worse. A good cure for the overshoot in circuit 6 is a series R-C element from the input and output of the current source to the joined base nodes. Values of 50 Ohm and 100pF cure the overshoot altogether. Circuit 7 does not show tendency to overshoot. Stabilizing the cascodes could be achieved by inserting emitter resistors to the cascodes with a value of up to 50 Ohm. Cascodes are great, but may oscillate and therefore, preemptive measures should be in place to prevent this.
This investigation has confirmed that different current sources behave very differently. There will likely even be an impact on sound due to temperature dependent bias drift of large parts of the amplifier. In case of the amplifier I built in year 2010, the current sources powering the input stage also bias the whole amplifier. For this project I chose type 2 due to the added safety that comes as a side effect of the strong negative temperature coefficient, resulting in lower OPS bias at higher temperature. The tendency to overshoot may also influence the sound. The current sources investigated here present a variety of solutions for different applications, each with its own unique properties.