Some Op Amp Audio Design Issues

A Ciphers By Ritter Page

Terry Ritter

2006 April 17


Some of the circuit details needed for good performance using low noise op amps are rarely discussed:
  1. An op amp used for audio or AC signals needs DC bias, both on input and output. That opens a whole raft of circuit possibilities, each with various tradeoffs.
  2. The usual bipolar supply does provide an intermediate bias level, but also generally causes the tiny input signal to flow through power supply filtering.
  3. The noise in a bipolar transistor op amp can be reduced with a low impedance signal source, but that imposes tradeoffs on the input circuitry.
  4. Large charged capacitors may need special discharge paths to protect sensitive op amp inputs.
  5. The op amp inputs may need to be protected against massive external transients, such as power on and power off of the signal source.

Power and Ground

The copper surface can be given any particular voltage. The copper represents one huge circuit node, and can save a lot of wiring if properly chosen. Normally, we would call the copper surface "ground," at least relative to that particular circuit.

Battery Power. The intent was to operate multiple shielded preamp stages, with the first stage amplifying microvolt signals. The strategy was to power each stage with its own 9V battery inside the shield. Batteries are particularly useful with preamps working on microvolt signals, because the alternative of an external supply would require penetrating the shield around each stage. With an external supply, substantial filtering would be required to keep outside signals outside, and extreme filtering is difficult at low audio frequencies. On the other hand, batteries themselves create noise (their output is the result of a chemical reaction, which presumably is subject to random molecular movement), so batteries also need bypassing.

Unipolar, Negative Ground. In most single-supply circuits, ground is the most negative voltage. For one thing, that makes power bypassing easy because each bypass capacitor can easily connect to ground. In a battery circuit, we may only need one battery.

Bipolar, Center Ground. Most op amp circuits use bipolar supplies: two equal-voltage low-impedance voltage supplies, with the level between the two supply levels taken as ground. With battery power, bipolar power generally means at least two batteries, although it is possible to create and use an intermediate reference level. With transformer power, making such supplies is fairly cheap and easy, but that generally means external power, which then becomes a filtering problem. With a bipolar supply, bypassing is still easy, although some capacitors will be oriented with (-) on the ground, and others will have (+) on ground.

A low-impedance intermediate supply level is extremely useful, despite the fact that few if any op amps have a ground pin. Output signals might need to sink or source considerable current to some reference level, and usually that is power supply ground. Input signals can be referenced to ground, which conveniently falls right in the middle of the op amp common-mode range.

However, the fact that op amps do not have a ground pin should raise a warning. One issue is the input signal circuit from the signal source, through the preamp, and then back to signal source ground. Assume that the input signal goes directly into the noninverting op amp input pin, which is no problem so far. The signal goes through the op amp and back out, typically through the (-) power pin, which is also no problem. But with a bipolar supply, to get back to input ground, the signal must traverse the negative power supply capacitor. When we have input signals moving through the power supply, that supply had better be very clean, and much cleaner than we might expect since op amps typically reject power supply noise.

Another issue is that a low-impedance input circuit is needed to minimize noise in op amps using bipolar input transistors. In the typical bipolar power design, the negative power supply filter capacitor is part of the input circuit. That means the filter caps may need to be larger than we might otherwise use simply to power a low-current op amp. If other units use that same supply, crosstalk could easily occur. Most preamp analysis does not expose how power supply design can affect bipolar transistor input noise levels.

Local Reference Ground. Another option is to create some intermediate voltage and use that for at least the output bias level. But we can make and use such a reference without necessarily treating it as ground.

Amplifiers Based on DC Power Require Bias

Circuits that process AC signals using DC power must rest at some intermediate voltage level or DC bias so that they may follow the signal whether it goes positive or negative. The bias levels can be different for input and output and often are.

Bias Means DC Blocking Capacitors

With a single-supply negative ground circuit, the input and output signal lines generally will have an above-ground DC bias voltage. This bias is typically isolated by capacitors used to conduct the AC signal and block DC.

One exception to the need for input bias is an N-JFET input, since the JFET gate can be at ground potential even in a unipolar negative-ground circuit. A similar exception occurs for rail-to-rail input op amps, which typically are CMOS devices, and also somewhat noisy.

Generally speaking, when we connect input or output to other devices, we will not know which side is most positive so we can properly orient a single polarized electrolytic capacitor between them. The usual solution is to use two electrolytics, one to isolate the input and one to isolate the output. Another approach is to create a bipolar capacitor by connecting two electrolytics negative-to-negative. In either case, cascaded boards would then use two caps to achieve half the capacitance of a single cap.

DC blocking capacitors raise various design issues, including cost and size and so on, but also more technical issues, like impedance. Every "DC blocking" capacitor is the action part of an RC filter consisting of a series capacitor, and a load impedance. These "high pass" filters attenuate low frequency signals. Making the filter "flat" requires the capacitive reactance at the lowest frequency of interest to be substantially smaller than the load impedance. It can take a big capacitor to deliver a low impedance at 20Hz. But using very large capacitors also generally means a very long turn-on charge time, and much more stored energy that needs to be respected and directed away during power-down.

Surprisingly, filtering goes both ways: Filtering the signals going into an amplifier is not the only result. In low noise designs using bipolar input devices, transistor noise is shunted away into the source, provided the source has a low impedance. For this shunting action to occur, it generally must occur through the input DC blocking capacitor, which now must be larger than one might expect. Fortunately, most noise energy is in high frequencies, so we can compromise somewhat, and may need to.

With all these blocking capacitor issues, one might wonder whether blocking capacitors could be avoided. Provided we know the source bias to be some reasonable positive level, one alternative is to not block it, but to instead accept and use that level. That can avoid having a capacitor in series with the input signal. The preamp then "amplifies" the DC bias by a factor of 1, while amplifying AC signals by much more.

Amplifying by different values according to frequency essentially requires each stage to be a filter, thus requiring and using the capacitor we just avoided. The problem with using the input DC bias is that very low levels of bias (e.g., 1.2V, 0.6V and, of course, a 0V ground reference) will be well outside the op amp common-mode range in a negative-ground circuit. While various modern op amps do support rail-to-rail input, and thus small ground-level input signals, that seems unlikely in low-noise bipolar op amps.

Input Noise

When a preamp has a high input impedance, it might seem that only a small input capacitor would be needed. A single film cap of about 1uF should do nicely. With that cap and a relatively high 100k input load, the low frequency response should extend down to about 10Hz. Unfortunately, that neglects noise generated by bipolar transistors when used as preamp input devices.

Typical low-noise work with bipolar input transistors depends upon having a low-impedance signal source to shunt and thus reduce noise developed in the input transistor. The transistor noise flows out the preamp input into the signal source output. That means the impedance between the signal source and the preamp op amp must be low at the frequency of interest, say, a tenth of the target impedance.

Since noise energy frequency response is typically flat, only about 5 percent of a 20kHz band of audio noise is below 1kHz, so we might live with a rolloff below 1kHz. The capacitive reactance of a 1uF film is about 159 ohms at 1kHz, which is pretty high if we want to shunt transistor noise.

So there are two limitations on the DC blocking capacitor value, and they are caused by the signal going one way and noise going the other through the same capacitor. For the signal, we want the capacitive reactance to be under a tenth of the impedance of the preamp input at the lowest signal frequency. That should be easy because the preamp input impedance should be high. For the noise, we want the capacitive reactance to be under a tenth of the impedance of the source at the lower useful noise frequency of 1kHz. That is harder because the source impedance may be low. Indeed, if it is not, there will be no advantage anyway.

Suppose we have a signal source of 50 ohms or less, and a preamp input of 100k. We want to handle signals above 20Hz, and noise above 1kHz. To shunt noise above 1kHz into 50 ohms, we might want a capacitor with reactance under 5 ohms, which means 30uF or more. That generally means a polarized electrolytic (or two 60uF electrolytics connected (-) to (-)). To shunt noise above 1kHz into 300 ohms, we might want a reactance under 30 ohms, or 5uF or more, which may be just barely possible with films. For signal above 20Hz into 100k, we want reactance under 10k, which means anything above 1uF. As an output capacitor, 1uF into a 10k load will be about 2dB down at 20Hz, and a higher load impedance will reduce that.

If we are willing to compromise a little, we could try a 3uF blocking capacitor (say, 3 film caps of 1uF each); this is a tenth of the value just guessed. With a 50 ohm source, simulation shows shunted noise down 3dB or so at 1kHz, which means about 95 percent of the removable transistor noise (1kHz to 20kHz) would be shunted away. We could move the cut down to 650Hz with a 5uF cap and raise the shunted value to almost 97 percent, but all this is a compromise anyway, and 1kHz high pass filtering seems very useful in this application. With a 300 ohm source, we are down about 3dB at 200Hz, which is nicer from a filter standpoint, if inherently less of a noise shunt. And in the signal direction we are flat above 20Hz.

Another approach to low noise is to use a transformer to change a low-impedance low-level signal into a high-impedance larger signal. That way the device input noise would not be shunted, but the larger signal would tend to make that less relevant. Unfortunately, input transformers which are flat across the audio band tend to be very, very expensive.

Energy and Input Protection

The larger the blocking capacitor we use, the longer it will take to charge up, given the very low current bias source we expect on a preamp input. With a large capacitor, the charging time constant could easily be seconds long, and cause problems at the human time scale. Even worse, as equipment is disconnected and re-connected, and powered on and off, the energy in the charged-up capacitor may become more than just a nuisance. The stored energy may, for example, find its way through the base of the input transistor, and change that device forever. Then we have to replace the op amp.

It seems reasonable to use Schottky power diodes to protect the signal input pin. A 1N5817 20V 1A (25A peak) should be conducting at least 2 amps before the input pin gets to even 0.5V above or below the supply, and at least 6 amps at 0.6V. Strangely, the higher voltage diodes in the very same series (30V and 40V) are less effective because they have higher turn-on voltages. The expected 200pF (0.0002uF) junction capacitance (per diode) is not an issue in this application, since we have at least 1uF from source output to op amp input.

Feedback Noise

Traditionally, op amps have had higher noise than most discrete designs. Now there are some good low-noise chips around (even though many are expensive and rarely found as surplus). When we use these devices, we add noise in the form of resistors (thermal or Johnson noise). But the added noise will not matter if we keep it below, say, a quarter of the op amp noise. Johnson noise varies as the square-root of the product of resistance and bandwidth.

  E(nV/SQRT(Hz)) ~ 0.129 SQRT(R)
  R ~ SQR( E(nV/SQRT(Hz)) / 0.129 )

     Ohms    nV/SQRT(Hz)   uV RMS (10k BW)   uV RMS (21k BW)
       1M      129.x           12.9              18.9
     300k       70.7            7.07             10.3
     100k       40.8            4.08              5.96
      30k       22.3            2.23              3.27
      10k       12.9            1.29              1.89
       3k        7.07           0.707             1.03
       1k        4.08           0.408             0.596
      300        2.23           0.223             0.327
      100        1.29           0.129             0.189
       30        0.707          0.071             0.103
       10        0.408          0.041             0.060
        3        0.223          0.022             0.033
        1        0.129          0.013             0.019

To get the expected RMS noise, multiply the nV/SQRT(Hz) value by the square root of the measurement bandwidth. Here, "10k" means 10,000Hz (so sqrt(10k) = 100); "21k" means 21,355Hz (so sqrt(21355) = 146).

For best noise performance it is important to keep the noise generated by the feedback resistors below the op amp noise specs. Actually trying to hear 1uV signals can be an education in shielding, grounding and thermal noise in the feedback path.

Inverting vs. Non-Inverting

There is reason to believe that the inverting configuration could have lower distortion than the non-inverting form. In the non-inverting configuration, the full input signal will occur at the (+) input, which then is matched with signal at the (-) input. Input level variation thus occurs on both inputs of the non-inverting configuration, which then depends upon a match between the input devices to minimize distortion. In contrast, the inverting configuration generally fixes a DC bias level on the (+) input, which is matched at the (-) input. Both inputs thus remain at a fixed voltage instead of a varying voltage, which avoids the distortion from working at different points in the signal range.

However, this input variation distortion may not be significant, especially with very small signals. And the inverting configuration also makes the source impedance part of the gain computation. That is particularly awkward when, as usual, the source impedance is poorly-known. The source impedance could even vary with frequency, thus producing a sort of random response instead of flat amplification. In contrast, the gain of the non-inverting configuration does not depend upon the source impedance. That makes the non-inverting configuration a more independent stage and thus a generally better interface to unknown external signals.

Terry Ritter, his current address, and his top page and electronics home.