This is part 3 in the series of posts discussing the (audio) measurement pre-amplifier project. In part 1 I’ve covered the motivation for this project along with the circuit schematic and detailed circuit description. In part 2, I have gone through the board layout consideration and showed the assembled boards. In this post, part 3, I will show some of the measurement results of the assembled boards. I will start with describing what it is I would like to measure, and how I plan on measuring it, including the limitations of the measurements I can make with the gear available to me. Then I will show the relevant result and discuss them.
The measurements I plan on performing can be split into 3 different groups. The first has to do with linearity of the pre-amplifier, to measure how much distortion it will have. Next are the noise measurements, as I want to verify the input referred voltage noise of the pre-amplifier to make sure it meets my target figures to allow measurement of low noise voltage regulators (and other devices). Finally are the “other” tests such as the accuracy of the True-RMS reading, the voltage limits of the output protection circuit, and so on.
I will start with the distortion measurements, and first describe the test setup and its limitation. Generally speaking, all my measurements are based on the EMU 0404 USB sound card. I have modified it somewhat to reduce distortion, as I’ve posted here. Which sets the limit of the distortion I can reliably measure with this setup. With sufficient tuning of the output and input amplitude, it can reach ~0.0005% THD (3rd harmonic dominant in balanced mode) throughout most of the frequency range of <10KHz. To extend this further, I have assembled a low distortion oscillator (posted here), and a notch filter with a similar frequency to extend the measurement capabilities of my setup to ~0.0001% at 1KHz. This is again somewhat sensitive to the input attenuation setting of the EMU, and can be as high as 0.0002% in some settings, as I’ve seen during experimentation. As far as THD+N is concerned, the EMU 0404 USB can only achieve ~0.002% (94dB) at best (full scale signal) in the 20Hz-20KHz band, whereas I except to the pre-amplifier to achieve ~16dB better than this.
These numbers set a limit on the minimum distortion (and THD+N) I can measure properly with my setup. Luckily, these are very low distortion limits, and are sufficient for every hobby use. For the noise part, this isn’t the case. However, I can partially circumvent this by measuring the noise with a lower input signal amplitude where I can increase the sensitivity of the pre-amplifier and extend the EMU capabilities this way. Another option would be to make use of the notch filter to achieve the same target. In the future I might consider upgrading the sound-card to extend these limits, but for the time being I see no need for this. More importantly, extending this significantly will cost much more than what the EMU was purchased for.
For all of the test shown below, I have used a 5-port 40W USB power brick that powered the pre-amplifier, the EMU, and the low distortion oscillator. As I’ve shared in the past, I’m really a fan of these USB power adapters, they are a great way of powering multiple small instruments/devices around the bench, and this is why I typically plan my builds for a 5VDC input.
First is a sweep of 20Hz-20KHz showing the distortion for the 2nd and 3rd harmonic, with the pre-amplifier in loopback mode. This means that the signal from the sound-card passes through the buffer (gain of X2) path of the pre-amplifier, then enters the input signal path (in 2V range, which has a gain of X1) and sent back to the sound-card. Since the EMU and the pre-amplifier both operate in BAL mode, I expect the even order harmonics to be lower than the odd harmonics. This is indeed what can be observed in figure 2, with the 3rd harmonic being much higher than the 2nd.
The distortion can be seen to be quite low, and is actually limited by the EMU 0404 USB which means this test won’t suffice to find the limits of the pre-amplifier.
To try an extend the limits of this test, I’ve connected the external low-distortion oscillator to the test setup. I’ve set the amplitude to 2Vrms which is the full scale value the pre-amplifier is meant to operate with in the 2V range, and attenuated the signal at the input of the EMU (after the pre-amplifier) until it is within the linear range of the ADC.
There are a few nice observation that can be found from this figure. First, the noise floor is on-par with what I’ve seen with no pre-amplifier in the circuit chain, which means the noise of the pre-amplifier is sufficiently low for my setup despite the use of the integrated line driver (as I’ve calculated in part 1). Next is the distortion, which is very low, even below what I would trust the EMU to measure, as I’ve mentioned earlier. Therefore, I’ve modified the test setup and placed the notch filter between the pre-amplifier and the EMU. The use of the notch allows a smaller tone to be used for the EMU, which in-turn shows up with less distortion of its own, extending the measurement range. First, I did this with no pre-amplifier in the signal part to try and get a reference measurement of this.
There are a few things to note here. First, the reduced signal amplitude at the EMU input because of the notch filter. Next is the increased noise, especially at lower frequencies. This is due to the high impedance of the notch filter at lower frequencies which increases thermal noise and makes it more sensitive to external coupling. After normalizing the harmonics based on the notch filter frequency response, we can see that we get just under 0.0002% residual distortion. This sets the practical limit of what we can measure.
Next, I repeated this test with the pre-amplifier added between the oscillator and the notch filter. This means that while the EMU will see a significantly reduced amplitude, the pre-amplifier will work with the full 2Vrms at its input. This is a single-ended (SE) oscillator, which makes it even worse in terms of even order harmonic distortion obviously.
We can see a few interesting things here. We see the 2nd harmonic is lower than what it was in the reference test (-117dBc vs -114dBc). Since the pre-amplifier sends out a BAL signal to the EMU even if its input is SE, this can explain why we see lower distortion here than in the reference measurement. However, this means that we cant say for sure if this -117dBc (~0.00014%) is generated by the pre-amplifier or the EMU. Next, we see that the 3rd harmonic is slightly higher than what it was in the reference measurement. However, when correcting for the transfer function of the notch filter, the 3rd harmonic is still at <-132dBc, which means its <0.000025% of the fundamental 1Khz tone.
Next, I re-did the same test, but this time I’ve disengaged the “GND input” switch of the pre-amplifier. Since the oscillator is floating, and sees a balanced impedance at the input of the pre-amplifier, this means the input is now an approximation of a BAL source. This will obviously make the measurement more sensitive to coupling of external CM noises, but for this test it is ok as we are interested in harmonics only.
Side note: the pre-amplifier was driving a balanced load through the balanced notch filter into the balanced input of the EMU when the “GND input” switch was in both positions. Driving a SE load with the pre-amplifier will degrade THD as the datasheet of the THAT 1606 shows, this is spec’d at ~3dB typical, but I’ve measured even higher reaching up to 6dB.
We can see both an increased noise coupling as expected, and that the harmonic distortion drops even further by an additional 3dB. Therefore I am confident that the distortion of the pre-amplifier, even at a 2Vrms input (maximum designed swing at this range) is <0.0001% at 1KHz.
This one is a bit of a problem to do properly now, due to the high gain and input impedance. The high input impedance makes it very susceptible to external coupling, while the high gain means that whatever couples into the input will be greatly amplified. Therefore, some of the experiments I did (such as measuring thermal noise of resistors) are fairly noisy due to external coupling.
However, there are some things that I can share about the input referred noise already. For instance, based on figure 3, we can see that the noise floor with the pre-amplifier in the signal path is unchanged compared to that of the EMU by itself. Calculating the input noise density we see in figure 3 (the -6dBFS input signal is 2Vrms), we see that it is almost at -150dB at a few KHz frequency, which is what I observe even with no pre-amplifier in the loop. This translates (with proper fft bin size correction) to ~300nV/rtHz. Since in part 1 of this series we have calculated the noise density of the pre-amplifier in the 2V range to be ~50nV/rtHz, it makes absolute sense that it will have minimal affect on the overall noise floor, which is indeed the case. This 50nV/rtHz figure is true for both input and output of the pre-amplifier, as its gain in the 2V range is X1.
Additionally, to try and measure the input referred noise density of the pre-amplifier on the more sensitive 2mV/20mV ranges, I’ve terminated the input with a 50ohm plug, and measured the noise density there.
We can see that despite the 50ohm termination, the fact it has such a high gain and no shielding (it is open on the work bench) means we can still see the 50Hz main coupled into the signal. In this measurement the 0dBFS is just under 4mVrms,which means this 50Hz tone is at 250nVrms. Calculating the noise floor (which is now a good 30dB over that of the sound-card due to the high gain) we can calculate it as being ~6.8nV/rtHz which is in good agreement with the calculated value of 7nV/rtHz.
In this section I would like to cover a few other tests that don’t translate directly into distortion or noise, but are relevant for general usage. For instance, the error of the T-RMS measurement, the sensitivity to common-mode (CM) voltage at the input, and the output protection.
T-RMS Measurement Accuracy:
To quantify the accuracy of the measurement I would have preferred to measure this across the entire supported voltage range. However, there are limitation at the extremes. At the extremely low voltages, external coupling could limit accuracy, while for the extreme high, I simply don’t have anything around the bench at the moment that could reach 200Vrms signal swing. In fact, I don’t have any amplifiers under work on the bench at the moment, so even 20Vrms is beyond what my function generator can create. Therefore, I will measure within the limits of what I have available here at this time, and I might expand this further in the future. However, since the difference between ranges is based on resistor values, I don’t expect significant difference between different ranges (1% resistors are used, so an error of 2% between ranges is the expected limit).
There are some observations that could be made by looking at this table. First, we can see that for lower frequency measurements, the gain errors between different gain setting are fairly small, and are as expected from the 1% resistors that are used throughout most of the circuit. However, as soon as we go to high frequency, we can see a significant error being added. Where is this coming from? For the high gain setting (2mV/20mV/200mV) we have a pole that limits the BW to 100KHz around the OPA1632. However, this should only result in <30% error (-3dB) at 100KHz, and only for the lower ranges. The source of the error at these high frequencies is actually the AD737 TRMS to DC converter IC. It has a high enough BW when operating with a small signal (albeit with some over-shoot), but since I am pushing it in these measurements to 0.5Vrms at the input of the AD737, it has a reduced BW. This BW will be higher (resulting in a significantly smaller error) when operating with smaller amplitude.
For audio applications this is obviously sufficient,as the errors until 20Khz are on the order of 1% (as expected by resistors tolerance). However, this is something that is easy enough to fix in a few different ways. It’s possible to attenuate this signal before the AD737, and increase the gain after it to compensate for this. Another solution is modifying the AD737 to use the low-Z input to extend the BW. The downside of this is that it adds an additional 8.2Kohm load (instead of the 100K it had originally) at the output of the OPA1632 that drives the TRMS->DC converter, but this is of minimal affect on performance so I’ve decided to go with this option. I’ve modified my pre-amplifier accordingly and repeated the measurements:
The results look much better now. The 100KHz shows an error of <30% (-3dB) on the 2mV/20mV/200mV ranges, just as expected by the 100KHz BW for these ranges. On the 2V range (and expected on the 20V/200V range too), the error is much smaller as the BW is increased significantly now. The same can be observed for a 20KHz square wave input of course.
Common Mode Rejection:
It is not uncommon to have some CM voltage at the input of the pre-amplifier. Ideally, this would be rejected completely. However, in practice, some of it will be translated into a differential voltage which will add to the measured signal. Therefore, we would like to maximize this rejection of CM. Since the most limiting part will be resistor matching, this can vary from one unit to another. Despite this, I’ve measured the remaining 1KHz tone at the output when a CM 1Vrms signal at 1KHz is fed into the pre-amplifier. The residual amplitude at the output of the pre-amplifier for each operating range is shown below. To get the input referred numbers one simply needs to divide by the gain of each range.
Output Voltage Limiting:
As seen in the schematic of the circuit in part 1 of this series, I have included voltage limiting (optional) at the outputs of the pre-amplifier. This is there as a fail safe to prevent damage to the DUT or sound-card/interface device if something fails. I think it is good practice to have this on an instrument which could potentially be subject to high voltages and other conditions that test instruments will come across. My unit has these limits implemented, and as can be seen by the THD measurements, for normal use, they do not pose any noticeable limit on linearity. To verify the limiting circuit is operating as expected, I’ve injected to the pre-amplifier a voltage that is over the maximum value for the expected range, and observed the output on the scope.
The voltage limiting can be seen to work as expected, with noticeable distortion starting at ~3V, and hard clipping at ~3.6V. This is for a SE output of course, and double the swing can be had for a BAL load. Again, as stated when reviewing the schematic, this is something that I think is highly recommended. However, it can be modified for a higher voltage (replacing the zener diodes) or omitted completely if desired.
In all the measurement I did up until this point, the circuit was driving the input of the EMU which is BAL, and therefore it was only driving 1Vrms on each of the outputs (for a 2V differential) signal. To find the limit of the clipping circuit, I’ve connected the pre-amplifier in loop-back mode, and enabled the GND-OUT switch (sw5 in the schematic). This means that the signal from and to the EMU is still differential, but the line driver now drives a SE signal into the input of the pre-amplifier. I’ve noticed that as I approach 1.8Vrms on the input, distortion starts rising to >0.001%. Therefore, with the parts I’ve used for the protection circuit it can only drive ~1.75Vrms SE or 3.5Vrms BAL before distortion from the voltage limit starts rising. This is acceptable by me, but I when I’m finished I will probably replace the 1SMA5914BT3G with the 1SMA5915BT3G. The will add additional headroom, and have a sharper knee to their transfer function, therefore they will extend the linear range to >2V(SE)/4V(BAL) while having less affect on the hard clipping of the output voltage.
Added on September 7, 2019:
I have replaced the limiting diodes with 1SMA5915BT3G as described above, and repeated the THD test vs amplitude. This modification is indeed sufficient to drive a 2Vrms SE (4Vrms BAL) signal without any distortion from the voltage limiter.
End of edit
Added on September 4, 2021:
I’m adding one more piece of information regarding the BW of the RMS meter (using low-Z input as described earlier), as a function of input amplitude. Due to the RMS->DC converter chip used, the BW will be a function of the input signal amplitude. I wanted to characterize this for the pre-amplifier so that its clear where practical limit of the circuit are. This is mostly important for noise measurements where the signal measured can be quite far below the full scale of the selected range, even on the lowest range. Figure 13 shows a few traces of normalized reading of the RMS meter for a few different input signal amplitudes. These measurements were all taken on the 2V range, where the BW of the pre-amplifier isn’t artificially limited to 100KHz as it is for the low voltage ranges (as Fig. 10 showed).
The figure shows a few traces from 30mVrms, up to 4Vrms. Just as a reminder, the pre-amplifier can work with 4Vrms (and even higher) at its input on the 2V range without much trouble. The 2V range is only called 2V, as this is where the default output limiting is for a SE connection. For a BAL connection the pre-amplifier can output 4Vrms as well. This can obviously be increased further by modifying the limiting circuits. The measurements shows in figure 13 were taken without any modifications to the circuit.
The things to note here are a few:
– For a 20KHz -3dB point (audio band), even 30mVrms is sufficient to get the desired BW.
– For a 100KHz -3dB point (power supply typical noise integration band), 100mVrms are needed.
– As amplitude is increased, gradually there is some peaking (~10%/1dB) around 300KHz.
– For higher amplitudes the BW goes far beyond the 100KHz point, and the pre-amplifier can be used as a generic RMS voltmeter up to a fairly high frequency.
Practically speaking, how does this translate to noise measurement on the lower ranges? Assuming this is used as a pre-amplifier for very low noise sources and we’d like to measure the integrated noise up to 100KHz, same as the pole that is added to the pre-amplifier on the higher gain ranges, 0.1Vrms equivalent reading is possible with minimal error. This translates to 10mVrms/1mVrms/100uVrms on the 200mV/20mV/2mV range. If you go below that, the reading will gradually increase in error. It is important to note here, that this is a limit of the RMS meter reading, not the actual signal path. Therefore, its always possible to connect the pre-amplifier output to an external RMS voltmeter, extending the BW.
End of edit
This is all for part 3 of this series. I think that overall the performance of the unit are great, and definitely meet my expectations and measurement needs.
The next step for this project will be the design of the front and rear panels to match the case that I will use. Once this is done and assembled, I will be able to revisit the noise measurements, which I hope will be less sensitive to external coupling once placed in an aluminium case.