A Simple and Flexible Audio Preamp Topology

A Variety of Essentially Similar Designs

xxx.

A Ciphers By Ritter Page


Terry Ritter

2005 November 5


Introduction

I wanted to measure shot noise. Shot noise is the statistical variation seen in huge numbers of discrete charges with independent launch times. As an audio level, shot noise is something like -120dBm, on the order of a microvolt. My (old) distortion meter will handle -60dBm, on the order of a millivolt. So I needed a gain of about 1000x or 60dB of very low noise, low level, audio band preamplification. In the end, that is not hard. But getting to that end can be somewhat more complicated than one might think.

A desire for low distortion and a device-independent gain value meant using negative feedback to set the gain. A desire to reduce power noise issues meant using battery power. (Although a battery is not noiseless, it does not have the usual 60Hz or switching component which then must be removed.) And battery power argued for designs with a single power supply and low current (say, about 2mA).

I wanted multiple, independent, and exchangeable gain stages so that I could swap them in and out, and put first whichever one had the lowest noise in my application. Since 60dB is a lot for a single stage, I decided on 30dB per stage (but if I were doing it again I might pick 20dB, which would make things easier for the op amp versions). Each stage would be shielded inside a "tea tin" of thin tinned steel, with battery power inside. There would be at most two connections through the shield box, audio input and audio output, both fully-shielded "RCA" audio jacks. The noise would be produced inside a "cookie tin" and then amplified by one stage, so that tin would need only an audio output jack.

The Investigation

My starting point was the literature on preamplifier design. The idea was to find a good, simple design, breadboard it up and use it. Originally there was going to be one breadboard. And when the first design did not work out, I just put the next design on that same board. When it finally became clear that this would be a research project, subsequent designs went on individual boards. Until I got at least a couple of designs with similar or at least related low-noise performance, I could not trust the results.

The resulting final setup used an OP27 op amp first stage, and an LT1884 op amp second stage. None of the discrete preamps really competed at the lowest signal level. Two stages of RC filtering were used to restrict the bandpass to above 1kHz (thus hiding 60Hz hum and 1/f noise) and below 14kHz (thus giving a 19kHz effective bandwidth). With this construction, I usually was able to measure a noise difference between no diode current and a 1uA current, a signal of about -114dBm, or about 1.5uV RMS. With the input shorted, the reading was about -127dBm, implying about 0.4uV RMS of preamp noise added to a pure low impedance source. That would seem to imply a noise floor of about 2.9nV/sqrtHz, or just slightly better than the typical 3.2nV/sqrtHz "Input Noise voltage density" specified by the maker, Analog Devices. I note that the OP27 is now considered obsolete, so better devices should be available, even within my somewhat arbitrary low-power constraints.

The Breadboarding Technology

My breadboarding approach is similar to but slightly different from what is commonly called "Manhattan style." My approach consists of first using a hand metal punch to manufacture 1/8-inch diameter "dots" or pads from 2-side PC material. It is convenient to first "tin" the board with solder on both sides, thus producing pre-tinned dots. Then I solder the pads to the board. That gives me solid yet insulated junction nodes or terminals. Parts and wires are then soldered to and between the pads. The common approach typically uses super-glue to position the pads, but sometimes it would not hold, and sometimes it was messy, and also the bottle did not keep long after being opened. So I tried soldering the pads, which worked out OK.

When soldering the pads in place, I try to put down several at a time, because that heats up the local area and makes soldering easier. Generally I brush liquid solder flux on the area to be used, then tin that area. Then, working on one pad, I hold it in position with a toothpick, and then bring solder on the iron tip to tack the pad in place. Since both the board and the pad are already tinned, the problem is mostly reduced to providing enough heat to melt both layers. If we get the area hot enough, the pad ought to suck solder into the volume between pad and board. It is important to use more solder than one might normally use for connections, because the solder mass is bringing heat to a copper surface which will distribute the heat quickly.

After a group of pads is tacked in place, I again hold a pad down with the toothpick, and bring enough solder on the iron to solder most of the circumference of the pad. Probably that is overkill, but that way the pads stay where they are, even under impact.

With this technology there is no need for a photographic layout, nor photographic production and drilling, nor the associated time cost. We just get some unetched board, and cut it to size using tin snips. The thinner "1 oz." copper layers is somewhat easier to solder than the thicker "2 oz." layers, but not a lot. For good appearance, I square up the sides with sandpaper on a flat surface, then round off all edges and especially the corners. Copper may be a "soft" metal, but it still can poke and cut human skin.

This breadboard technology has various advantages:

In terms of problems, there is some tendency to get very fine solder hairs from pad top to ground, thus the need to protect the battery. But a battery should be protected in any development technology. When pad shorts happen, they are easy to find (just ohm each pad to ground) and repair (use solder braid to remove excess solder).

Prototype Board 1

[]

The first try was a preamp based on the LM394 low-noise dual "transistor." This was from National Semiconductor AN-2222 (1979), Fig. 4, "Ultra Low Noise Preamplifier." The biasing system is at least unusual, making modifications difficult.

A bipolar input device generally requires a low output impedance from the source or previous stage. That is in parallel to the base-emitter and emitter resistances, and so tends to hide Johnson noise in the device and design. Now, an (ideal) input capacitor has no Johnson noise. But the input level still divides across the capacitive reactance and the total bipolar device and emitter resistance. With a low input resistance we need a big input capacitor, and then we need to wait for it to charge up after power-on.

In the end, my prototype had stability issues.

On the same board was the second try: a preamp based on one device in an LM837 low noise quad op amp. It was surprisingly noisy.

Prototype Board 2

[]

The next try was another LM394 preamp, with an LM837, based on Analog Devices MAT02 data sheet, Fig. 24, "Low-Noise, Single-Ended x1000 Amplifier." This seemed promising, but did use two input transistors, which had to be more noisy than one. Moreover, it required the usual bipolar op amp supply, which was less convenient. As a result, however, the input was direct-coupled and ground referenced, thus avoiding the usual input capacitor.

With hindsite, now I would use a different op amp, and then perhaps a different LM394 or even some other bipolar, and would consider a 60dB stage gain in this particular case. In the event, however, my prototype had stability issues, so I just moved on.

Eventually, I did get a much different version to work. I used on op amp section to simply buffer the collectors, and another as an integrator "servo" to drive the emitters. A 2.2k input resistor gave -4mV on the input which, in the end, I thought was too much. Also a dual ferrite bead was needed on the base to only partially tame instability. Perhaps I should have been more suspicious of the LM394 or LM837.

Prototype Board 3

[]

For try number four, Google searching web groups led to a preamp design by Win Hill (one of the authors of The Art of Electronics). Having just built two previous LM394 versions that had problems, this seemed like a good thing to try.

One potential problem is the differential stage, since two transistors have got to be more noisy than one (square root 2 or 1.414 times).

My prototype was both unstable and noisy. Possibly the noise was a result of oscillation.

In many of these prototypes, I had heavy radio-frequency-interference (RFI). Although the ultimate application would have substantial shielding, during run-up testing, each board would be on the bench without much around it. Nevertheless, there is a lot more "dirt" at microvolt levels than one generally suspects using millivolt instrumentation. 

From: Win Hill 
Subject: Re: Super-Low-Noise Microphone Preamp, Advice needed
Date: 2000/08/10
Message-ID: <3992B930.98D5BB6E@mediaone.net>#1/1
Newsgroups: sci.electronics.design
[...]

The discrete transistor solution I posted a few days ago for you
equals the performance of the opamp circuits, but draws only 2mA
from a single 9V battery and works to 6V end-of-life.  OK, here's
another one, fresh off mt drawing pad, this time with a dual BJT
input instead of a JFET.

[...]
           ,----+-----+--------+------+------+------- +9v
           |    |     |        |      |      |        BATT
           |  1.10k   |        |     470    75k
           |    |     |        e      |      |
          30k   +---------+- b   Q2   +--||--+ 2.2uF
           |    |     |   |    c pnp  |      |
           |    |     |   |    |      c      |
        ,--+    |     |  47pF  +--- b        |          R10
        |  | Q1 |     |   |   5.6k    e  Q3  |  G = 1 + --- = 50
      1.0M | LM394CH  |   |    |      |      |          R9
        |  |    |     |   '----+------+      |
    0.1 |  |    c     c               | Vx   |    1 uF
 In -||-+---- b         b -+--,    ,--+-+--------||---+--- Out
           |    e -+- e    |  |   R10   |    |        |
   Zin    27k      |      47u 1M  2430  c    |        |
   1meg    |       c       |  |    |      b -+        |
   20pF    +---- b   Q5    '--+----'    e    |       47k
           |       e       R9 |      Q4 |  diode      |
           c       |        49.9    ,---+    |        |
        Q6   b ----+          |     |   |   4.7k      |
           e       |        470uF 22uF 1.0k  |        |
           |      453         |     |   |    |        |
  GND -----+-------+----------+-----+---+----+--------+-- GND

 The amplifier should have an input noise level of about 1.5nV/rtHz.
 [...]

Prototype Board 4

[]

For try number five, the bipolar input Win Hill article led to a somewhat simpler design based on a JFET. Since the FET gate was a ground referenced signal input, this design also could avoid an input capacitor in some cases. Normally, however, the preceeding stage will have some bias, so one capacitor generally is needed. But we do not need both an output and an input cap in the same signal path! Unfortunately, unless we know the bias voltage levels for the preceeding stage output and this input, we may not know which direction a polarized capacitor should face. That argues for a high input impedence, which allows a smaller value non-polarized cap (film, not ceramic). And the JFET design has that high input impedence.

With a lot of parts removed, and some other modifications, eventually this became the basis for a number of prototype boards, using various types of input device. Perusing the literature will reveal a variety of designs with similar or related topologies. 

From: Win Hill 
Subject: Re: The definitive guitar -> soundcard preamp.
Date: 2000/07/24
Message-ID: <397C3DC1.9C420342@mediaone.net>#1/1
Newsgroups: sci.electronics.design
[...]

OK folks, here's my improved preamp circuit.  It will have very
low distortion due to an unusual design feature: each device in
the signal chain operates at a constant current, independent of
output voltage.  Consider, that's better than class A operation!
In addition it has negative feedback.

[...]
                   ,------+--------+----+------- +9v
                  R2      |        |    |
                  2.7k    |       330  7.5k
                   |      E        |    |
                   +--- B          +-||-+ 10uF
                   |  Q2  C     Q3 |    |         (R8+R9) R10
                   |  pnp |        C    |  G = 1 + --------- = 20
           2N5457  |      +----- B      |           R9 * R8
           etc.    |     R3        E    |
             Q1  |-'     2.7k   Vx |    |   1 uF
    In ---+----->|      ,-+--------+-------||---+----- Out
         |      |-,   R10      Q4 |    |       |
          |        |   1.8k        C    |       |
          |        +----+            B -+      47k
         10M       |    |          E    |       |
          |     R8 |   R9          |  diode     |
          |    ** 750  100     ,---+    |       |
          | select |    |      |  680  1.5k     |
          | for Vx |  100uF   22uF |    |       |
   GND ---+--------+----+------+---+----+-------+----- GND

[]

Board #4 was the first acceptable implementation, which is why it still exists to be shown (the others do not). But it has a lot of stuff I eventually discarded.

On this board, power comes in at the upper right, through a battery-protect resistor (47 ohms), and a reversed-voltage protection transistor (a P-channel MOSFET used in reverse). The clips short out the 2.7k resistor, thus removing Q3. That provided much better RFI isolation, presumably by reducing loop gain. In later designs, I kept Q3 but removed Q4 and replaced it with a resistor. I also removed R3 and R8. In many cases I was able to actually hear a noise reduction when resistors were removed.

Summary

Overall we have a particularly clean and simple audio preamp topology elaborated in many versions with some common goals:

Beyond that we have some ideal goals, not always reached because something more important prevents it:

A whole range of different versions were built and tested:

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

Last updated: 2005-09-25