QA472 Measurements

[matt]

QA472 Mic Preamp Analysis

The QA472 is an upcoming mic-preamp product to replace the QA471. The QA471 was based on an SSM2019, while the QA472 is based on the INA849. The INA849 offers much tighter gain limits, improved THD, and similar noise performance as the SSM2019. The mic pre offers, 0, 10 and 20 dB of gain, which are selectable using a front-panel button that cycles through the gains.

The QA472 also has an ultra-low-noise JFET amp with 30 dB of fixed gain.

The purpose of this post is to show how to make the measurements to characterize a mic-preamp. As the product being evaluated here is preliminary, certain specs may change prior to shipping.

QA472 Front Panel

The front panel of the QA472 is shown below.

PRE1: Mic Pre

From left to right, there are indicators for Power (the QA472 is powered by USB) and also +48V phantom. PRE1 has both BNC and XLR inputs. A push button to the left of the BNC will enable system phantom generation (+48V). To the right of the BNC inputs is an input select button, and LED indicators for BNC or XLR inputs.

To apply Phantom Power, the +48V button must be pushed and held for 1 second. To turn phantom off, a momentary push is needed.

Phantom is never applied to the BNC inputs–only to the XLR inputs. When you switch from XLR to BNC inputs with phantom power active, the phantom will be turned off and a short “bleed down” delay will be imposed before switching to the BNC inputs to ensure the full phantom voltage isn’t applied to whatever you have connected to the BNC inputs.

The XLR female inputs do not have 1/4 TRS, but a push-lock tab is provided. To the right of the XLR input is the gain select button, and LEDs to indicate the selected gain. And finally, a single-ended BNC output.

PRE2: JFET Pre

The JFET pre-amp is based on the TI JFE150 “ultra low noise” audio N-Channel JFET. This part has the ability to deliver noise performance about 5-10 dB better than the OPA1612 opamp with low (<1K) source impedances. And 20-30 dB better noise performance with higher-Z sources.

The QA472 front-panel settings are remembered through power cycles. So, if the QA472 is part of a measurement setup, a power cycle will restores the measurement settings you had in place. The QA472 is not USB controllable.

The measurements below were performed on REVC boards, with mods made that will be reflected in a REVD board, and REVD is expected to go to market.

Measuring PRE2: JFET 30 dB Preamp

JFET Amp Gain

A first measurement we’ll make of the JFET pre is the gain, because we’ll need to know this number for the “referenced to input” (RTI) noise measurements we make next. With the QA403 L+ output connected to the PRE2 Input, and the PRE2 output connect to the QA403 IN+ (and with IN - shorted), we can make a gain measurement as shown. Note that with 30 dB of expected gain, we’ll set the generator to -40 dBV.

Above we can see the amp delivering 29.89 dB of gain. This is a “no cal” amp design based on the TI app note for the part. TI’s overall circuit is a hybrid, using a JFET front end, and a precision opamp to close the loop. More units need to be measured to understand the spread of the JFET amp’s gain. But initial looks show them to be fairly tight and in good agreement with spice.

JFET Amp Noise

With the gain known, we can short the input of the amp. Because this amp is high-gain and also very low noise, it’s important to short the amp input with a true 0 ohm shorting block instead of the often used 50 or 75 ohm shorting block. Below we can see the thermal noise in a 20 kHz bandwidth of various resistor values. For example, a 120 ohm resistor normally has a thermal noise (in a 20 kHz bandwidth) of -134.03 dB. And with 30 dB of gain, we’d expect that to result in an amp output noise of -104.3 dBV. As we’ll see below, with this gain and noise, even a 120 ohm resistor will degrade the performance of this amp.

With the input shorted, we can see the A-weighted spectrum:

If we replace the 0 ohm shorting block with a 75 ohm shorting block, there’s about 1.5 dB of degradation:

The QA471 had a 30 dB amp based on a OPA1612 low-noise audio amp. The RTI noise of the OPA1612 is -129.42 dB. So, the JFET delivers about 6 dB better noise floor over the OPA1612. And as the source impedances increase beyond 1k, the JFET performance will get better and better due to to the current noise in the OPA1612.

JFET Distortion

Below is a sweep of input level versus THD. We can see at -20 dBV input (10 dBV output) that clipping is starting to occur.

Looking at the spectrum with -40 dBV input we see the 2H is about -105 dB below the fundamental.

Note the THD+N is -84 dB, which is considerably worse than the THD. We can see the N-D (noise minus distortion) is -94 dB. But now, let’s short the input again with a 0 ohm shorting block. Note the N-D figure improves about 10 dB, which means the THD+N is limited by the noise in the output of the QA40x analyzer, not the preamp. From this, we can roughly estimate the THD and THD+N are both about -103 dB at -40 dB input.

JFET Frequency Response

The plot below shows the frequency response of the JFET amp. The amp is bandwidth exceeds the bandwidth of the QA40x in 192Ksps sampling rate.

In the next week or so, the mic preamp measurements of the QA472 will be posted. The last board had some stability issues related to excessive capacitance due to switching topology on the INA849 gain setting resistor that will hopefully be resolved in the next board pass.

[matt]

Measuring PRE1: INA849 0/10/20 dB Preamp

PRE1 on the QA472 has both BNC and XLR inputs. The XLR inputs can be connected to the 48V phantom via 6.8K resistors, but the BNC inputs cannot. Because the large 6.8K resistors, the phantom voltage isn’t a threat to the either the inputs or outputs of the QA403. This is because the 6.8K resistors will limit the current to about 5 mA or so. So, you can freely plug any of the inputs or outputs on the QA472 into the inputs or outputs on the QA403 for short periods without concern for damage.

INA849 Gain Tolerance

There are a lot of choices for mic preamps, including the SSM2019 and THAT1510. And all are great for noise. Where they fall a bit short, however, is gain tolerance. The SSM2019 allows you set a fixed gain with a single resistor. And for a gain of 10, you can see the from spec sheet below that the typical gain error is +/-0.2 dB. But it could be as bad as +/-0.5 dB.

And the THAT1510 is similar in that the gain error can be as much as +/-0.5 dB:

Now, when you are offering the user a constant gain range from 0 to 60 dB via a pot, the error is inconsequential. But for measurement with specified fixed gains, the error needs to be much better than 0.5 dB. The INA849 is a new laser-trimmed diff amp offering gain errors of +/- 0.1%, which is +/-0.009 dB. Quite a difference compared to the +/-0.5 dB barn door of the older generation devices.

Of course, using 0.1% resistors will double that.

QA472 Gain Error

The INA849 gain setting resistor pins are very sensitive to stray capacitance, and beyond 10pF or so (per SPICE) it can result in instabilities. The gain resistors are switched by photomos switches which have been selected for their low capacitive loading and also the off-state capacitive coupling.

The gain resistors used are as follows:

R (Ohms)	Theoretical Gain
inf	0 dB
2.8K	9.94 dB
665	20.01 dB

Given the limits on the INA849 gain tolerance and resistor tolerance, we’d expect the gains of a random unit to be very close to the above.

At 20 dB gain setting we see:

And at 10 dB gain setting we see:

And at 0 dB gain setting we see:

The above are a considerable improvement over the QA471, which offered a single fixed gain with a much larger tolerance.

Pre1 Noise Measurement

As with Pre2 above, we’ll first measure the noise at 0 dB gain with A-weighting applied. This yields -104.30 dBV output noise, often called RTO (referenced to output).

And at 10 dB. This yields -103.91 dBV output noise (again, RTO). If we wanted to reference to input, this noise would be -113.91.

And at 20 dB, we have -102.42 dBV. And referenced to input, this would be -122.42 dB.

By specifying the gain in the dBV context menu, we can make an input-referred measurement of the 20 dB gain where the calculation is done for us automatically:

And we can then enable the phantom power, and see check for degradation in noise performance. There is none. Phantom power noise can come from several sources, and is often related to power-savings modes (such as pulse skipping) used on DCDC converters.

PRE1 Distortion

Let’s sweep the distortion from -50 to 0 dB on each of the gains: 0 dB, 10 dB and 20 dB. To do this, we’ll use the AMP THD versus InOut Level Automated Test with the settings below.

The resulting plot is as follows.

This plot is likely bumping up against the limits of the QA403, primarily on the QA403 DAC side of things. The INA849 has a THD figure typically around -127 dB. However, that has been degraded somewhat by the capacitive coupling in the photomos switches used to do the gain switching on the INA849.

What’s Next?

Next up will be some analysis on gain spreads on the QA472 Pre1 and Pre2, along with a look at using a notch to investigate the THD limits of the QA472 Pre1.