What is baudline?
Mystery Signal
Digital Loopback 16

vendor SigBlips
product baudline signal analyzer 1.01
interface digital loopback
duplex full
channels 2
resolution 16 bits
max rate 15+ MSample/sec, CPU limited
operating system Linux x86_64 2.6.12-1
driver none
test date Aug 24 2005
notes Enable the tone generator loopback in the Input Devices window.  Set the digital gain in the tone generator to 0 dB.

This card is part of the Full Duplex DAQ comparison survey.

The mission of this digital loopback test is to measure the performance of baudline's Tone Generator and decimation filters without the analog influence of a sound card.  These measurements define the baseline level of quality for 16-bit resolution.

Sample Rate
The following table of measurements were made using the sample rate stability technique.  The base rate column divided by the decimation column is the sample rate column.  Since this is a digital loopback there is no ADC/DAC clock.  The loop error column uses a tone generator loopback method for a high accuracy measurement of the relative difference between the ADC and the DAC clocks.  This test is a measure of the tone generator's phase accumulator stability coupled with the sensitivity of baudline's Hz measurement window.  This error can be corrected and is discussed below.

base rate decimation sample rate loop error
48000 1 48000 +0.0993 PPM 
48000 2 24000 +0.0662 PPM
48000 4 12000 +0.0883 PPM
48000 8 6000 +0.0207 PPM
48000 16 3000 +0.0336 PPM
48000 32 1500 +0.0376 PPM
48000 64 750 +0.0336 PPM
48000 128 375 +0.0431 PPM
48000 256 187.5 +0.0452 PPM

The loop error has an accuracy of more than 0.1 PPM.  In real world measurements this accuracy can be improved by a couple orders of magnitude if baudline is run in a dual channel mode.  One channel being the analog loopback signal and the other coming from the digital loopback.  Then use the frequency measured from the digital loopback instead of the frequency dialed into the tone generator to calculate the PPM error.  Think of this as a runtime calibration that corrects the error of the dialed in tone generator output.

Frequency Domain
Baudline was wired up with the Tone Generator digital loopback in full duplex fashion.  Since this is a digital loopback link and the performance is independent of sample rate only the base rate of 48000 will be used.  The sample rate will be varied by using baudline's decimation setting in the Input Devices window.  The following sections show a brief overview of the channel frequency response and the distortion metrics as a function of decimation rate.

The spectrogram image is of a linear sine sweep that excites all of the channel's different modes.  The orange curve in the Average window is collected WGN.  Both are an application of the swept sine vs. WGN technique and are equivalent measures of the frequency response.  The green curve in the Average window is the loop backed sine wave signal used by the ENOB window and the other distortion metrics. 

There is no noise floor crosstalk curve for two reasons.  First, because the test signal is mono and the crosstalk measurement requires at least stereo channels.  Second, because digital silence is absolute and the noise floor response is at -inf dB which is off the bottom of the display..

48000 -nodither

The purple curve is a 1800.00 Hz sine wave and it shows the effect of having no dither.


48000 / 2

48000 / 4

48000 / 8

48000 / 16

48000 / 32

48000 / 64

48000 / 128

48000 / 256

By looking at the above plots a couple interesting observations that relate to decreasing sample rate can be made.

The spectral lobe width of sine sweep increases as the sample rate is lowered.  This is because the swept duration is 10 seconds for all of the test runs.  This is effectively increasing the sweep speed which widens the spectral lobe..

The energy of the orange WGN curve drops by 21 dB from the decimate by 2 to the decimate by 256 case.  That works out to a 3 dB drop for each halving of the sample rate.  This is also known as decimation gain.

The variance of the Average spectral curves grows as the decimation ratio increases.  This behavior is due the collection duration remaining constant.  A lower sample rate means less samples for a given duration.

The following table of measurements were made using the technique described in the sine distortion application note.  It is a full duplex test that uses a loopback of the tone generator to measure the various distortion parameters.

48000 -nodither  +99.88 dB  -106.45 dB  +99.02 dB  +16.154 bits  +115.27 dB 
48000 +96.91 dB -103.54 dB +96.06 dB +15.662 bits +112.73 dB
48000 / 2 +95.29 dB -102.96 dB +94.61 dB +15.421 bits +111.45 dB
48000 / 4 +96.40 dB -101.91 dB +95.32 dB +15.540 bits +111.46 dB
48000 / 8 +96.44 dB -102.51 dB +95.48 dB +15.567 bits +111.44 dB
48000 / 16 +96.42 dB -104.62 dB +95.81 dB +15.621 bits +113.55 dB
48000 / 32 +96.68 dB -102.81 dB +95.73 dB +15.608 bits +110.00 dB
48000 / 64 +96.66 dB -103.93 dB +95.61 dB +15.638 bits +113.42 dB
48000 / 128 +96.72 dB -103.21 dB +95.84 dB +15.627 bits +109.50 dB
48000 / 256 +96.36 dB -105.27 dB +95.83 dB +15.625 bits +109.75 dB

There is no quantization distortion because the loopback is a pure digital channel.  The Histogram window below is a perfect example of how a white Gaussian noise distribution is supposed to look.

The sample rate stability loop error slowly improves as the decimation ratio increases but only by a factor of 2.  Theoretically the PPM error should be improving by a factor of 2 for every doubling of the decimation rate due to the increase in bin resolution.  This isn't happening so other internal algorithmic numerical noises must be the limiting factor.

The distortion measurements are fairly equal at all of the different decimation ratios.  This validates the correctness of baudline's decimator algorithm.  There is a 0.2 bit drop in ENOB from the decimation off to the decimate by 2 transition.  This ENOB anomaly improves slowly in the following decimate by 4, 8, and 16 transitions.  Not sure what it means but it could be filter imperfections causing the initial drop and then gradual rise could be due to internal decimation gain.

The distortion measurements are a couple dB higher and the ENOB is 0.4 bits greater when dither is disabled.  Dither eliminates spectral spurs at specific frequencies (see purple curve above).  So dither is a compromise.

These distortions and errors are extremely minor when compared to the measurements from a full duplex analog sound card.

A pure digital signal was used as a loopback test source and baudline operated as expected.  There were no obvious flaws, glitches, spurs, or distortions.  This test validates the integrity of baudline as a signal generator and analyzer. 

This test of the digital loopback also creates a baseline of performance that all of the other sound cards can be compared against.

Copyright © 2005 - group - blog - site map