Soundcard Noise Performance
The post looks at the noise performance of some popular/high-end sounds cards for the PC. One is the EMU0204, which is a cost-effective USB device built off the AK4396 ADC. The second is the EMU1616, an older PCI-based sound card aimed at the pro market. The EMU1616 was built using the PCM1804 ADC.
The performance of both of these cards are then compared to the QA400.
The EMU0204 is currently a favored device for those wishing to make PC-based audio measurements. The cost of the EMU0204 is extremely competitive, and the noise performance is extremely good. Our aim with the QA400 is to offer a comparable level of performance, include a competitive audio analyzer program with that, and bring the two together at a price well below what can otherwise be achieved today using alternate avenues. In addition, we can remove much of the uncertainty inherent in PC-based approaches today, while at the same time offering a level of automation and programmability that hasn’t been see at this price before. .
Unless noted, measurements on the EMU1616 and EMU0204 were made using ARTA at 48Ksps, 24-bit with ASIO drivers. The QA400 measurements, unless otherwise noted, were made at 192Ksps, 24-bits with drivers specific to the QA400 application.
Input Clipping Levels
The purpose of this test is to understand where the input stage begins clipping. Based on the captured spectrum, the LED indicators are reliable indicators that clipping is occurring.
Right Line Input: With the input gain at 9 o’clock position, the input clipping LED turns on around 1.4Vrms. With the input gain at the 3 o’clock position, the input clipping LED turns on around 22 mVrms.
Left Line Input: With the input gain at the 9 o’clock position, the input clipping turns on around 1.6Vrms. With the input gain at the 3 o’clock position, the input clipping LED turns on around 15 mVrms.
The slight variation between left and right units is likely due to slight difference in the knob positions. The EMU0204 is advertised as having 60 dB of variable gain in the front end, and it appears well suited to dealing with a wide range of input signals.
Note that for precise audio measurements, the variable inputs can be both a blessing and a curse. The variable input gain is great when you want a wide range of equipment to be measured. However, every time the knobs are adjusted, the system performance changes and the system must again be re-characterized to understand the performance limits. This can be very time consuming. Additionally, as will be shown below, at the higher gain settings (eg past 2 o’clock) the noise floor begins to rise quickly.
ADC Clipping Levels
The purpose of this test is to see how close to 0 dBFS the clipping indicator illuminates. Ideally, the ADC will clip well before the analog stage will begin clipping.
Right Line Input: A 1Vrms 1 KHz signal was input into the right line input, and the input knob was adjusted counterclockwise until the clip LED extinguished. At this point, ARTA noted the peak was –0.5 dBFS.
Left Line Input: The test was repeated in the left line input, and the ARTA noted the peak was –0.3 dBFS.
In summary, there is good agreement between when the input clipping LED illuminates and when the ADC clips.
To look at noise floor, the measurement mode was switched to dBV. A 178Vrms 1 KHz signal (-15 dBV) was input into the left and right channel. The input gain was changed to 0.5. This will allow us to use the input gain to correctly set the reference to 0 dBV. On this particular 0204, this required the output level to be at –8.5 dB when using the internal generator.
Right Input Noise Floor: To achieve –15 dBV reported level, this required the input knob at roughly the 10 o’clock position. With the –15 dBV level set, the spectrum appeared as follows using the internal generator driven from the headphone output.
And if we turn off the generator (leaving the output connected to the input) the following spectrum remains. The spurs we see here are power line (fundamental and perhaps 3rd harmonic). Note the cable is from EMU0204 output to input, so most likely is that this is output noise that is being measured here.
Next, we can actually unplug the input to the right channel. This yields the following very clean spectrum:
For the left channel, we can repeat the same experiment.
Unplugging the left channel input yields the following noise floor. Note there is a little bit of leakage displayed at 1 KHz and 2 KHz.
Turning of the generator and unplugging the input yields the following. Remember, this is with the input knob at roughly the 10 o’clock position.
If we turn the knob to the minimum position (around 7 o’clock) we get the following
And if we turn the knob the 4 o’clock position we get the following. In short, the EMU0204’s noise floor is highly dependent on the location of the input gain knob. And low gains the noise floor is quite good, at high gains it’s quite poor. From a pure numbers perspective, this makes sense as the front-end gain will also gain up any noise ahead of the input op-amp and including the noise inherent to the op-amp.
In the plot below, notice the noise at 1 KHz is about 40 dB higher at high gain versus low gain.
Switching to 192Ksps, we can see the spectrum noise floor is reasonably flat to 40 KHz, at which point the sigma-delta noise shaping begins.
Using ASIO drivers, 24-bit ADC and 48Ksps, the left input was set to the 10 o’clock position, and input clipping occurred at around 550 mVrms. Here we see the clipping point is aligned very well with the 0 dBFS point.
Changing to dBV and adjusting the input signal to –15 dBV and then adjusting the input gain to report –15 dBV (it’s a very sensitive adjustment, so this isn’t quite exact) yields the following.
With the input removed, we see the following noise floor. While the noise is quite reasonable out at 1 KHz, the noise is fairly high at the lower frequencies. Recall the 0204 was better than –140 at 20 Hz.
With the sample rate changed to 192Ksps, the due to noise shaping is again seen.
The internal generator on the QA400 was set to generate a –15 dBV signal, the reading confirmed below.
The input was then removed to reveal the noise floor of the QA400. Note the internal leakage can be seen at 1 KHz.
The generator was then turned off to eliminate the leakage. This reveals the ultimate noise floor of the QA400, and as can be seen the measurement in a 20 to 20 KHz bandwidth is –105.1 dBV. Note this is for 192Ksps, 32Kpoints.
The test was repeated at 48Ksps with the averaging increased:
For quick comparison, the EMU0204 at 192 KSps shown below
And the EMU0204 at 48 Ksps is posted here again (this is the same shot as above).
Finally, the extended spectrum on the QA400 is shown below. Here, we can see the small differences likely due to decisions by the ADC designers. In the CS4272 (the part used in the QA400) the spectrum is flat to about 45 KHz and the noise bumps to about –108 at 100KHz. In the AK4396, the bump starts a bit earlier, but the height of the peak is quite a bit less.
Measuring Sansa Clip+ on the QA400 and EMU0204
The Sansa Clip+ is an ideal device for PC reference measurements because it has great audio performance and it’s battery powered. This means we can better isolate the noise performance of the PC-connected hardware without having to worry about ground loops potentially clouding the measurements.
For this test, the Sansa played a –10 dBFS 1 KHz signal and the output volume was adjusted to give 177 mVrms output. This –15 dBV signal was then used to calibrate the EMU0204 to the same absolute levels at the QA400.
Below is the Sansa signal on the QA400, correctly showing –15 dBV input level. Notice the FFT noise floor is around –115 dBV or so at 192Ksps
Repeating the test at 48 Ksps shows the reduced noise floor dropping to below –120 dBV
Next, we can look at the Sansa playing into the EMU0204 at –15 dBV output. For this test, the Sansa is back to 48Ksps and the volume control is around 10 o’clock position (pre-amp gains are set to 0.1). The plot above is the ideal comparison here, since both devices are at 48Ksps and both device are using a 32K FFT. There are a few differences to note: The noise around 20 Hz is higher on the QA400. It’s virtually identical other places, however. Note the 3rd harmonic on the QA400 is 10 dB higher, however, the EMU0204 has a lot of spurious components from 100 Hz up to 10 KHz that aren’t there on the QA400.
Checking again at 192Ksps to ensure an equal comparison yields the following. Here we see the noise floor at around –112 dBV and the second harmonic at –101. This is very similar to QA400.
This post looked at the noise performance of an older, very high-end EMU1616 PCI device, a recent bargain EMU0204 device and the QA400. The EMU1616’s performance relative to the EMU0204 and QA400 is lacking, showing just how far budget desktop performance has come in the last few years.
At identical settings, the QA400 and EMU0204 show almost identical input performance when the EMU0204 is set to its lower gain ranges. At the higher gain ranges, the EMU0204 noise floor rises fairly quickly, limiting its utility as a low-noise device.
The QA400 also offers symmetric input channels, while the EMU0204 has different input structures for the left and right channel which will certainly result in measurement differences.
Finally, the QA400 offers precise and calibrated fixed gain performance, eliminating the need for frequent knob adjustments. The QA400 also bypasses the Windows Audio Subsystem, allowing you to be certain you are getting the promised performance every time.
A future examination will compare the output performance of the QA400 and EMU0204.