The composition and characteristics of the 32-bit floating-point ADC system --- ZOOM UAC-232 USB audio interface evaluation

TestGuru

The composition and characteristics of the 32-bit floating-point ADC system --- ZOOM UAC-232 USB audio interface evaluation [Copy link]

 

download attach  save to album

2023-10-11 14:37 上传

In reality, it is extremely difficult to achieve a dynamic range of 125 dB with a single ADC, even if it claims a bit depth of 24 or 32 bits, because it is difficult to keep the noise floor below 1 bit with such a large dynamic range. Today, microphones can have a dynamic range of up to 140 dB, which matches the dynamic range of human hearing. Therefore, it is a technical job to adjust the input analog gain so that the main part of the audio signal is within the dynamic range of a single ADC. If the audio signal is too large, the waveform will be clipped, while if the audio signal is too small, it will lose its fidelity and even be buried in the noise floor (see the figure below).

Problems caused by the relatively limited dynamic range of a single ADC

Rather than pushing the performance of a single ADC to its limits, it is better to find a different approach. Current audio devices that support 32-bit floating-point formats usually have two ADCs working together and synthesizing a single 32-bit floating-point data stream output (see the figure below). The "low-gain" ADC is optimized for strong audio signals, while the "high-gain" ADC is optimized for weak audio signals. When the high-gain ADC clips due to excessive signal amplitude, the low-gain ADC does not. When the low-gain ADC cannot clearly capture the sound above its noise floor due to the low signal amplitude, the high-gain ADC still has enough margin above its noise floor. Therefore, with this dual ADC configuration, the total dynamic range can be significantly extended, achieving low-noise and non-clipping recordings, even without adjusting the analog gain. This is why the Zoom UAC-232 does not have an input analog gain switch and knob on the panel. One of the main goals of this article is to find out the actual dynamic range that the device can achieve. Another focus is to explore its ADC switching algorithm and its impact through experiments.

Dual ADC architecture supporting 32-bit floating point format

Multi-Instrument is one of the first test and measurement software to fully support 32-bit floating point format ADCs. In 32-bit floating point mode, the software always uses the raw sample value of 1 as 0 dBFS. In the software, you can enter the actual voltage value represented by this value through [Settings]>[Calibration]>"Other/ASIO">"Range" to achieve calibration. It should be noted that in 32-bit floating point mode, 0dBFS does not necessarily represent the maximum level value. The maximum level value where clipping distortion occurs needs to be measured here. In the following tests, unless otherwise specified, the "Range" is set to the default uncalibrated 1Vp. The advantage of this is that the measured instantaneous voltage value can directly reflect the original sample value, for example: an instantaneous voltage value of 1V indicates that the original sample value in 32-bit floating point format is 1.

The following tests were performed using the ZOOM UAC-232 ASIO Driver V1.1.0 released on May 30, 2023 and the ZOOM UAC-232 Mix Control V1.0.0.13 released on February 9, 2023. The ZOOM UAC-232 Mix Control is used to control the mixer settings inside the ZOOM UAC-232. It sets different default digital gains for different input devices (XLR microphone input, LINE input, and HiZ input) (see the table below). Modifying these default values will automatically save them in the hardware device. Please note: After modifying the digital gain settings, the aforementioned "range" calibration will be invalid and must be recalibrated. In 32-bit floating point mode, the quality of the recording will not change as the digital gain changes. Therefore, in the following tests, the digital gain is always kept at 0dB.

TestGuru

2. XLR microphone input

download attach  save to album

2023-10-12 12:24 上传

Maximum level without clipping at XLR microphone input (digital gain = 0dB )

Maximum level with clipping at XLR microphone input (digital gain = 0dB )

download attach  save to album

2023-10-12 12:28 上传

A-weighted noise floor level for XLR microphone input (digital gain = 0dB)

Unweighted noise floor level for XLR microphone input (digital gain = 0dB )

Therefore, the equivalent input noise floor level can be calculated as:

This result is consistent with the -127dBu in the ZOOM specification.

download attach  save to album

2023-10-12 12:33 上传

SNR of XLR microphone input at -60dBFS (digital gain = 0dB )

TestGuru

download attach  save to album

2023-10-14 02:17 上传

THD , THD+N , SINAD , SNR , Noise Floor Level for XLR Mic Input at -1dBFS (Digital Gain = 0dB )

download attach  save to album

2023-10-14 02:19 上传

SMPTE IMD for XLR microphone input at -1dBFS (digital gain = 0dB)

DIN IMD of XLR microphone input at -1dBFS (digital gain = 0dB)

CCIF2 IMD for XLR microphone input at -1dBFS (digital gain = 0dB )

download attach  save to album

2023-10-14 02:30 上传

Bandwidth of XLR microphone input at 48kHz sampling frequency ( digital gain = 0dB )

Bandwidth of XLR microphone input at 96kHz sampling frequency ( digital gain = 0dB )

Bandwidth of XLR microphone input at 192kHz sampling frequency ( digital gain = 0dB )

download attach  save to album

2023-10-14 02:32 上传

Crosstalk at XLR microphone input at 1kHz (digital gain = 0dB )

TestGuru

download attach  save to album

2023-10-16 22:45 上传

Noise floor level for XLR microphone input vs. input RMS amplitude (calibrated) (digital gain = 0dB)

Once again, the noise floor level measured at the low end of the signal amplitude in the above figure is slightly higher than the 134.3dBFS measured previously because the output noise floor level of the signal generator used is relatively high. As can be seen from the above figure, at a signal level of about 0.03Vrms (equal to -34.9dBFS), the noise floor level suddenly rises. Since this is still within the minimum range of the signal generator output, the output range switching of the signal generator is not involved. Therefore, the sharp rise in the noise floor level is obviously caused by switching from a high-gain ADC to a low-gain ADC.

XLR microphone input THD vs. input RMS amplitude (calibrated) (digital gain = 0dB)

The figure above shows that the THD of the high-gain ADC initially decreases as the input signal amplitude increases until the signal amplitude reaches about 0.01Vrms. Then the THD starts to increase until the signal amplitude reaches about 0.03Vrms. After that, the low-gain ADC is switched on, and its THD begins to decrease as the signal amplitude increases until the signal amplitude reaches its turning point of about 0.7Vrms, at which point the THD starts to increase again. As can be seen from the figure, the lowest THD that the two ADCs can achieve is about 0.0006%.

THD+N vs. input RMS amplitude (calibrated) for XLR microphone input (digital gain = 0dB)

The figure above is a superposition of the effects of the two previous figures. At a signal amplitude of 0.03Vrms, the switching between the two ADCs is more clearly shown.

download attach  save to album

2023-10-16 22:51 上传

Overall input-output linearity graph for XLR microphone input (digital gain = 0dB )

In order to see in detail whether this linearity is affected when the ADC switches, it is necessary to reduce the step sweep range. The following figure shows the measurement results of 300 steps from 0.0141Vrms to 0.0424Vrms. It can be seen in the figure that there is a small jump at 0.03Vrms. This is exactly where the ADC switching occurs as found in the previous test.

Input-output linearity plot of an XLR microphone input near the ADC switch (digital gain = 0dB )

To further explore the small "non-linear" jump caused by the ADC switching in the above figure, it is necessary to reduce the sweep range again, this time from 0.028Vrms to 0.032Vrms in 300 steps. The test results are as follows. As can be seen from the figure, the height of the jump is about 0.0003V.

Close-up of the input-output linearity plot of the XLR microphone input at the ADC switch (digital gain = 0dB )

TestGuru

download attach  save to album

2023-10-19 12:20 上传

Impulse response of an XLR microphone input involving only the high-gain ADC (digital gain = 0dB )

Impulse response of an XLR microphone input involving a low-gain ADC (digital gain = 0dB )

Since the duration of the impulse signal is extremely short, the difference between the sampling points of the response signal is extremely large, which is enough to cover up any difference in gain and zero offset between the two ADCs. Therefore, we cannot expect to see obvious ADC switching traces on the waveform. The spectrum of the impulse response is the frequency response. From the two figures above, we can see a little difference in the frequency response of the two ADCs measured by this method.

Zoom in vertically on the two waveforms above (as shown in the two figures below) to observe the possible changes in the noise floor level before and after the impulse. As expected, the noise floor level remains unchanged in the first figure. In the second figure, the change in the noise floor level clearly reflects the switching between the two ADCs: from the high-gain ADC before the impulse to the low-gain ADC during and after the impulse, and 60ms later, it returns to the high-gain ADC again due to the weak signal.

Impulse response of an XLR microphone input involving only the high-gain ADC (digital gain = 0dB )

Impulse response of an XLR microphone input involving a low-gain ADC (digital gain = 0dB )

download attach  save to album

2023-10-19 12:26 上传

Linear sweep to trigger ADC switching (digital gain = 0dB )

When zooming in on the image above to examine the waveform at the ADC switching point, no noticeable jumps are observed. The waveform appears fairly smooth, as if no ADC switching has occurred. The image below is a close-up of the transition region from the high-gain ADC to the low-gain ADC. The top (or bottom) of the sine wave is an ideal location to examine the noise level because it is relatively flat. In the image below, the third peak of the sine wave clearly has a higher noise level than the first two peaks, which further confirms the ADC switching position indicated by the red amplitude envelope. It can also be observed from the red amplitude envelope below that the noise level increases gradually rather than instantaneously, which means that there is a DSP algorithm for splicing in the transition region to fade in the new data stream from the low-gain ADC and fade out the old data stream from the high-gain ADC to achieve seamless splicing. The transition region length appears to be about 0.5 milliseconds.

Transition region where the ADC switches (digital gain = 0dB )

TestGuru

download attach  save to album

2023-10-21 11:31 上传

Advantages of 32-bit floating-point ADC

Although there is inevitably a very small misalignment between the high-gain ADC and the low-gain ADC, the switching between them is very smooth, and no jump traces are found in the waveform, thanks to the DSP splicing algorithm used in the transition area. Moreover, this transition is completed in a very short time and hardly causes any audible traces, and the error introduced to the sound measurement is completely negligible. The dynamic range of the XLR microphone input of ZOOM UAC-232 is about 134.3dB or 136.4dBA, which is more than 10dB higher than those solutions using a single ADC. With the appropriate microphone, it is possible to capture the sound within the entire human hearing range without sacrificing the fidelity of weak signals and without causing high-intensity signal clipping distortion, and it eliminates the trouble of manual analog gain adjustment. This is an extremely attractive technology in the sound recording and measurement industry.