A Good-enough Microphone Preamplifier (Part 1)

Introduction

This is the first part of an article about building a "good-enough" microphone preamplifier. It outlines some of the design choices and initial research.

Motivation

As I mentioned in an earlier post, I acquired some microphones a while back to use for recording live concerts. The microphones need phantom power, but I would also like to be able to use them with a digital recorder or indeed with a PC. None of my kit is of "professional" quality, but is good enough for its purpose - recording groups of performers often in less than ideal venues. There are, of course, plenty of commercial products that would solve the problem to a greater or lesser extent, but since building a homebrew microphone preamp is something of an electronics rite of passage, I thought I might have a go, not aiming at the ultimate in low-noise, high-bandwidth performance, but for something "good enough". The first hurdle was to decide what "good enough" might be, the second to work out how to check if the solution met the specification.

Requirements and Constraints

Phantom power was clearly a necessity, along with good (common mode) noise immunity to deal with longish cable runs amidst electrical noise. A reasonably level frequency response in the audio spectrum and enough low-noise gain to bring the condenser microphone signal up to (PC) line level would be needed too. Preferably, I would like the whole unit to operate from USB power (or from a powerbank): 5V power is now ubiquitous - though a number of different supplies (including the phantom power) would likely have to be derived from it.

Phantom power poses a number of design problems - there's a much-quoted paper from THAT Corporation which discusses many of the issues - and subsequent papers discovering that the problems were even worse than initially stated. For this reason, I wanted to use a transformer to decouple the input from the 48V supply: it doesn't eliminate all the problems, but it does reduce them. Unfortunately, decent audio transformers can be very expensive (and, conversely, cheap transformers can be dreadful), but I found a reasonably priced candidate, the Monacor LTR-110, which I thought it might be worth investigating. The first step, then would be to check out the transformer.

It would also be necessary to check out power supply solutions to see if they could provide the necessary voltages and currents with low levels of noise and ripple. 

This would all require appropriate test equipment.

Test Gear

Decent audio test gear is expensive and largely beyond most hobbyists' pockets. However, there are software options that use a PC sound card (assuming it's of reasonable spec) as both a signal generator and measuring device. They have the potential to be very accurate (for example using 24 bit samples and a 96kHz sampling rate), but the output and input levels of the sound cards is uncalibrated and the quality of the buffer amplifiers may not be great. The digital oscilloscopes that fall into the affordable range may have a wider bandwidth and be properly calibrated, but have much lower sampling range (8 bits, typically, and likely no more than 12) which means you can potentially "see" more but any measurements are likely to be rather imprecise. 

Digital oscilloscopes are, though, of more use in testing power supplies, particularly switching supplies that operate at high frequencies.

After some experimentation, I settled on ARTA for audio performance measurement as it gave more consistent and reproducible results than some of the alternatives I tried. An entry-level Picoscope proved sufficiently capable for the observations required outside the 20Hz-20kHz range.

Transformer Performance

The transformer manufacturer quotes a frequency range of 15-30,000 Hz which is more than adequate for the purpose envisaged. However, before measuring the performance of the transformer, it was first necessary to measure the loopback performance of the PC sound card to see how much colour it might be adding to the measurements.

The frequency response cut-off at 20kHz is a function of the measurement software - ARTA doesn't appear to support sampling frequencies above 48kHz, but otherwise the frequency response is very flat.

Similarly, based on a 1kHz tone, the card's noise performance is reasonably good with little sign of harmonics distortion.

With these measurements as the baseline, we can then see the results for the transformer interposed between the PC's line input and line output. As is typical of transformers, there is some third-harmonic distortion present, though it is at a low level.

 

It's instructive to compare the frequency response measured by ARTA with that measured on the oscilloscope (signal generator sweep limited to 40kHz):

 

You can see that the transformer response is only around 5dB down at 40kHz.

 

From these measurements, it seems that this transformer will be suitable for its proposed use.

Power Supply Performance

There are a number of cheap switching buck/boost converters available from online suppliers, usually based on the XL6009 chip and they are obvious candidates for producing both the split-rail power supply likely to be required for the preamp itself and for the 48V phantom power. 

I first looked at this board, which provides a (variable) positive and negative rail from a single rail supply - at rather higher currents than will be required for this application.

 

Initially, I hooked up a 9V battery and, using a multimeter, set the output to 12V-0-12V. The XL6009 switches at up to 400kHz, so the relatively small smoothing capacitors ought to be able to produce a clean supply.

Using the oscilloscope to observe the power output, there is a small amount of ripple of 3-4mV on the supply rails:

The scope is indicating that the ripple frequency is variable - it's not quite clear if there's a problem measuring it reliably at low amplitude or whether the converter is "pulse skipping" because of the low load. The ripple is probably acceptable as it is, but I wanted to see if it could be reduced further without too much complication or expense. I tried adding a larger smoothing capacitor and inserting a series choke. The results can be seen below:

As can be seen, a 100μH choke brought the ripple under 1mV, a better result than a 1000μF capacitor.

One thing to note is that it isn't necessarily clear how much noise is attributable to the power supply and how much due to environmental noise picked up by the scope probe - particularly when switched to its "x10" range. Some brief testing suggested that in fact this level is higher than expected and the actual performance of the power supply is better than the initial measurements would imply. The noise results are more than "good enough" but I may revisit the measurements once a prototype has been constructed.

For the phantom power supply, I looked at this module (and took a little more care with the measurements):

 

There are quite a few superficially-similar modules to be found from Internet vendors - and YouTube reviews boasting of high output voltages, however most are specified for a maximum output of 35V - and their switching diode is a 40V part which will not survive higher voltages for very long. The manufacturers of this board claim a maximum output voltage of 50V and the switching diode is rated at 60V/3A. There's also an output choke (marked 2R2) which is missing from the "lesser" boards. The output capacitor appears to be rated for 50V only, which leaves little safety margin, but it should be possible to replace it with one with more headroom - if the board is capable of providing the 48V we require (or close to it).

Loading the output with a 6k8 resistor, a voltage of 48V could comfortably be achieved both from a 9V battery and from a 5V power supply. Using the battery (to eliminate switching noise from the 5V supply), the noise output from the power supply was as follows:

The switching spikes are visible at a frequency of around 185kHz, but the overall noise level is relatively low. Using an addition 220μF on the output, the noise level was reduced below 1mV:

In this case, the addition of a choke (alone) caused more ringing on the switching spike; the choke in combination with the additional capacitor only marginally reduced the noise level compared with the capacitor alone.

The conclusion is that these power supply boards would likely be suitable to generate the required power supplies from a 5V source, though possibly some additional filtering may be necessary.

Candidate Designs

So, with the knowledge that a suitable transformer was available and that the power requirements could be met, it was time to look at some candidate designs for the preamplifier. Discrete designs offering good performance are likely to involve a high component count, but there are IC-based solutions that can potentially offer very good performance with relatively few additional components. Although it would be possible to use IC solutions operating with a single 5V power supply (given we're only driving a consumer grade line input), the availability of cheap voltage converters offers more options and the chance for the design to have wider applications, so for the present moment I'm content to assume we have a split-rail 12V supply available.

The first option is a simple op-amp based circuit (in this case using half of an NE5532), the second uses a THAT1510 which, though aimed at the audio market, is basically an instrumentation amplifier. The two designs are shown in the following diagram:

In the top circuit, the 200R resistor roughly approximates the DC resistance of the transformer secondary, balancing the bias currents to the op-amp. The transformer secondary is connected to the (high-impedance) non-inverting input of the op-amp. In the lower circuit, the differential inputs are biassed with a single resistor connected to the centre tap of the secondary meaning a higher resistor can be used than the 1510 datasheet advises as any noise appears in common mode across the inputs. These are the minimum circuits that can be constructed for each device.

The performance of the op-amp circuit was measured as follows:

 

For the Instrumentation Amplifier, the corresponding measurements were:

Although these measurements put the op-amp circuit marginally ahead, one of the problems of PC-based measurement is that the result vary according to the level of input and output signals and these are difficult to accurately calibrate. On balance, I'd say that within the limits of repeatability there's not much to choose between them. Out of interest, the 1510 circuit frequency response measured using the scope with its wider frequency range shows a response that is down only 2dB at 60kHz:

 

It seems that either design would be more than adequate for the task. However, a practical solution will require adjustable gain - and both of the designs will see shifts in the DC offset as the gain is altered and potential consequent audio artefacts even if the output is AC-coupled. It would also be good for the final device to provide some sort of metering.

In the next part of the article, we'll look at the practicalities of taking a basic breadboard concept to the prototype stage.