Amplifier input protection

The good, the bad and the ugly

Abstract

Amplifiers inputs are highly sensitive and should be protected from ESD and high voltage in general. ESD occurs all the time handling any equipment outside ESD controlled environment. Overvoltage events may be caused by connecting a high level source, maybe even an amplifiers loudspeaker output. This topic seems to attract surprisingly little attention. The standard recommendation for operational amplifiers is to clip the input to the supply rails. For a high power amplifier, this is often not really feasible since the rails may be over 100V. A small signal transistor typically has an absolute maximum VEB rating of 5V to 6V and this is the voltage the input protection needs to clamp the signal to. In this article, I will investigate and compare three different circuits clamping performance and impact on distortion using Spice simulation. All models used for simulation in this article are from Bob Cordell and are considered trustworthy by the audio community.

Three different clamping circuits

I simulated different circuits to see how they perform.

Reverse biased diodes

clamping circuit using reverse biased diodes
This is the first circuit I came up with. It is based on the idea to clip to the rails. Low voltage supply rails are formed by resistors from the high supply rails together with Zener diodes and filtered with small capacitors. The clamping diode is backwards biased during normal operation. Clamping occurs at 5V. The good thing about this arrangement is that unlike real power supplies, which most often struggle with excess voltage injected, the Zener diodes just conduct any current to ground. Actually this is a shunt regulator. The capacitor together with the resistor from the amplifiers supply rails forms a low pass filter with a low corner frequency, suppressing noise from the amplifiers rails, plus shunts Zener diode noise. If the amplifier is powered off, the circuit mostly behaves like the Zener or TVS diodes circuit instead.

Forward biased diode stack

clamping circuit using stack of forward biased diodes
I simulated this circuit just for fun and out of curiosity to see what happens. Ten diodes in series are somewhat ridiculous, but in HiFi sometimes strange things may happen. The idea behind is that the reverse biased diodes nonlinear capacitance of ten diodes in series could be very low. The circuit makes the signal clip very softly at 5.5V.

Zener and TVS diodes

clamping circuit 3
Circuit 2 is far too complicated and consumes way too much PCB real estate. So the most obvious thing to do, is to use Zener diodes in series with their anodes or cathodes tied together. This is the smallest of all three solutions. This diode circuit likely performs somewhat similar to TVS protection diodes offered by many companies for many different interface applications. Clipping occurs softly at 4.4V. This circuit was also investigated using a bidirectional TVS diode. I chose diode ESDM3031MXT5G because it was the only one in this voltage range with a Spice model available and is advertised for audio use.

Clamping waveforms

THD

THD was simulated at 2.5V, which is roughly half the clipping voltage. This is also the voltage that makes the amplifier clip, so it is a worst case scenario. For acceptable performance of the protection circuit, the THD should be well below the amplifiers distortion.

Reverse biased diodes

The good. With the second harmonic at -100dB and a THD of 0.0007% at a level where the amplifier would clip at the power rails, there is not much to complain about. This should be mostly inaudible.

Forward biased diode stack

The bad. Here the second harmonic is at -70dB and with a THD of 0.022%, this becomes audible and may even dominate the THD of the whole amplifier.

Zener and TVS diodes

The ugly. Second harmonic at -55dB and THD of 0.123%, this circuit could be used as an effect but not as input protection for a HiFi amplifier.

Zener diode vs. ESDM3031MXT5G clamping circuit FFT
I also simulated FFT for diode ESDM3031MXT5G, which on paper seems perfect. Maybe a bit too perfect to be true. At 3.3V reverse voltage, leakage is specified 100nA maximum. The Spice model however shows only 3nA leakage at this voltage, which seems not uncommon for TVS diodes and the Zener diode even has 110µA leakage in simulation. To me it makes sense that leakage is lower since TVS diodes are designed to be used below their breakdown voltage whereas Zener diodes are designed to be used at their breakdown voltage. According to the model, breakdown occurs extremely sharp from 260nA at 4.8V to 3.5mA at 5.2V. Distortion in simulation is very low and suddenly increases dramatically. I find the sudden increase of distortion a bit unrealistic. A very high and slanted noise floor also indicates there could be something special with the model. Maybe the diode is good indeed, but the model is probably not. I bet that this diode does not behave even close to this on the bench.

THD vs. level

I compared THD vs. level for all circuits presented including the one using the TVS diode:

Here is a more detailed view of the circuit using reverse biased diodes and the one using ESDM3031MXT5G.

Clamping performance

So how well do those circuits clamp ESD events at the input? I simulated contact discharge using the human body model.

Voltage Reverse biased diodes Forward biased diode stack Zener diodes
2kV 6V 16V 8.3V
4kV 6.9V 24V 9.7V
6kV 7.7V 32V 11V
8kV 8.5V 40V 12.4V

The reverse biased diodes and Zener diodes perform best here while the diode stack fails completely. None of the circuits can keep the voltage below 6V for high voltage ESD. Anyway, having the voltage clamped to 8.5V is better than brutal 8kV firing into the super sensitive input transistor. What first surprised me is that the reverse biased diodes clamp at lower voltage than the Zener diodes despite being mostly the same circuit. Looking at the current wafeforms, it becomes obvious that the capacitors conduct a the majority of the peak to ground, which helps digesting the peak a lot.

Clear winner of the clamping contest would be diode ESDM3031MXT5G. For ESD pulses between 2kV an 8kV, the clamped voltage would not exceed 5.7V to 6.7V. Maybe this is even realistic. However, I don't trust the model of this component.

Permanent input overload

In case the amplifiers input is permanently connected to a high level source like another amplifiers speaker output, the reverse biased diode circuit and Zener diodes circuit clamp the voltage to 6V and the diode stack to 9V. I simulated the case a 200VPP signal is connected to the input. The diodes are fine with roughly 100mA peak current. The 1k resistor will dissipate 4W. So this part is either sized accordingly or accepted to burn down happily and given some space to surrounding components in order not to ignite them while burning down. Increasing the resistance comes with a noise penalty and maybe isn't desirable.

Protection of the loudspeaker output

The amplifier interfaces to the outside world with the loudspeaker output as well so overvoltage protection should be considered for this interface. Actually, the loudspeaker output has more scenarios to consider to protect against. The most likely scenario is overload or even short circuit, but this is complex to deal with and deserves a extra article. Many articles have been written and I won't write another one. In this article, I focus on overvoltage only.

Voltage injection

In case a voltage source is connected to the loudspeaker output and the amplifier is powered on, the amplifier would start to fight the injected voltage to zero the output. Damage depends on which one is stronger. In case the amplifier is powered off, injected voltage would be conducted to the supply rails via the catch diodes in BJT output stages or the parasitic body diodes in MOSFET output stages. This would then charge the power supply capacitors until the amplifier comes to live, what then results in aforementioned scenario.

ESD

In case of ESD, the output transistors are protected by those OPS catch diodes in any case. Amplifiers that use negative feedback are still vulnerable to high dV/dt injection at the output in any case since the inverting input is connected to the output and the input is highly sensitive. So it seems necessary to implement a protection strategy for the inverting input. The same extra rails from the reverse biased diodes protection circuit could be used and only two additional diodes are required for clamping the inverting input to those rails. I simulated the impact on THD using my amp under development and cannot find any deviation between unprotected and protected inputs. However, added capacitance at the inverting input may impact stability. In case other feedback techniques are used, like error correction, a different strategy is required since the input in this case is floating with the loudspeaker output.

Conclusion

Among the circuits simulated using trustworthy models, it is quite clear that the reverse biased diodes circuit performs best in all cases. This is what I'm going to use for my next amplifier on both inputs. It seems to work reliably using standard components that don't add anything exotic to the BOM and can be simulated with certain confidence. Simulation suggests that the TVS diode outperforms this solution in all aspects, but this may not be true in reality.

Below schematic shows an example application of the reverse biased diodes clamping circuit in a simplified LTP input stage. Note that the clamping circuit needs to be placed after the DC coupling capacitor in order to avoid asymmetric clipping. The example also shows two anti-parallel diodes between the amplifiers inputs. This is pretty standard in operational amplifiers. The anti-parallel diodes may help protecting the input, but may also destabilize the amplifier by adding capacitance to the inverting input. For amplifiers using negative feedback, the voltage difference between the inputs is close to zero. During gross overload there might be a voltage difference between the inputs and the diodes start conducting. I've observed this may be causing the amplifier to start ringing. Use with care.

The TVS diode could be a stellar performer or just a case of a badly designed Spice model. TVS diodes in general are similar to Zener diodes, although optimized for different parameters. I don't buy that the TVS diode in question is that perfect. The only way to find out is to put it on the bench and measure it. I did some research and found out that the typical macro model of a TVS diode is designed in a questionable way. ON Semiconductors AND8254 explains how such a macro model is designed. The key is that it contains a voltage source in series with the diode used to model breakdown. This voltage source backward biases the diode and this is the explanation why the diode shows such a sharp breakdown characteristic. This also explains why the characteristic shows striking similarity to the reverse biased diodes clamping circuit, which also contains backward biased diodes. A real TVS diode of course has no voltage source built in and therefore likely behaves differently. The TVS diode will instead show behavior somewhat similar to to the Zener diode clamping circuit, although it is not clear whether performance would be better or worse than with Zener diodes.

Appendix: Zener diodes vs. TVS diodes

Writing this article and observing unexpected effects in simulation made me research a bit more about Zener diodes and TVS diodes. In this appendix I collect some of my research.

Wikipedia about Zener diodes
[...] However, many diodes described as "Zener" diodes rely instead on avalanche breakdown. Both breakdown types are used in Zener diodes with the Zener effect predominating at lower voltages and avalanche breakdown at higher voltages. [...]
Diodes.com application sheet (AN1142)
Diodes AN1142 Image © by Diodes Incorporated, used here under "fair use" regulation
[...] Both devices rely on the Zener breakdown effect and have I-V characteristics that look very similar. [...] Although they appear similar, a closer look at the I-V characteristics in Figure 5 immediately shows the differences. The Zener Diode (blue trace, Vz) has a much sharper knee, steeper slope and tighter voltage tolerance (dashed lines and shaded region) than a TVS device. These differences come from design and optimization [...]