MicroMPX is a codec that is designed specifically for FM, to transmit the full MPX spectrum from a studio to one or more transmitter sites. Our aim was to create a codec that uses as little bandwidth as possible, without sacrificing quality. When we started development, we were targeting bitrates around 500-700 kbit/s. However, extensive listening tests have convinced us that even at 320 kbit/s, there's no perceivable difference in the audio signal. This blog post contains measurements of the signal at different bitrates.

If you want to know what the benefits are of sending the full MPX spectrum instead of just left/right audio, please read The advantages of composite clipping.

Virtually lossless quality

At high bitrates, differences between the original MPX signal and the signal after decoding are virtually nonexistent. We can measure the difference signal (subtracting the original MPX signal from the decompressed MPX signal) which is typically below -100 dB, with a maximum of -85 dB. The images below show the spectral view of the original signal, and the difference between the original and decoded signal (encoded at 800 kbit/s).

No audible degradation at low bitrates

When we designed MicroMPX we had a number of requirements:

1. No perceivable effect on the audio.
2. All the advantages of composite clipping, perfect MPX peak control and asymmetry in the L-R area must be maintained.
3. No difference in reception, which includes:
   1. Pilot quality, and protection (no noise around it).
   2. RDS signal quality
   3. Channel separation
   4. Stokkemask (ITU-R SM.1268): Compliant signal in = compliant signal out

While MicroMPX's compression is technically lossy, it uses techniques that don't cause typical problems. It does not have the pre- and post-ringing or watery sounds that are present in most lossy codecs such as MP3, AAC, OGG etc. MicroMPX is designed to be incapable of encoding MPX signal peaks above 0 dB, which means it cannot cause overshoots. It attempts to keep the waveform as close to the original as possible, instead of just keeping the sound as close as possible, protecting it against overshoots and reception issues when the signal has been optimized for ITU-R SM.1268 (Stokkemask).

MicroMPX never removes any sound or audio details: it's not a psychoaccoustic codec that removes hard-to-encode soft sounds that you're supposed not to notice. It can only add minor amounts of noise where the original signal already contained similar frequencies, making it inaudible. Any FM reception hiss would mask the additional noise it even further.

Lowering the bitrate does not affect the RDS, pilot, peak control and channel separation.

Comparing codecs and bitrates

Lowering the bitrate in MicroMPX only affects the audio bands (L+R and L-R). It has no effect on the RDS and pilot sections. This means we can look at the demodulated audio to determine what the effect of lowering the bitrate is, and to compare its audio quality to that of other codecs.

Let's look at some measurements of the difference between the original and compressed/decompressed audio. Note that these differences might not be noticable when listening to the full compressed/decompressed signal. Listen tests with several subjects showed that artifacts in MP3 are clearly audible at 128 kbit/s and to a lesser extent at 320 kbit/s, while they aren't audible in MicroMPX at 320 kbit/s.

Interpreting artifact levels

Psychoacoustics tells us that loud noise can mask more subtle sound. Soft noise and artifacts are inaudible, until they reach a level where they suddenly become very noticeable. Small differences in artifact levels can cause them to be clearly audible or not at all. Normally, a 1-2 dB difference doesn't have a big impact, but as soon as audio is masked by other audio, changes of this magnitude can have a large impact on our perception of audio quality. This is the case for any codec: not just MicroMPX, but also MP3 for example.

The difference signal of audio that has been through the MP3 codec has pre-ringing, sounds rougher and contains tones that were filtered out in its psychoacoustic stage. MicroMPX does not contain these undesirable characteristics, but its difference signal sounds similar enough to that of MP3 to make comparing their levels useful.

As you can see in the graph below, at 320 kbit/s, MicroMPX adds artifacts that are 6 dB lower than those of MP3. That's the total difference over the entire spectrum, at the frequencies where the artiacts are most noticeable the difference is even bigger - around 12 dB. These artifacts decrease in power by 6 dB, every time we step up MicroMPX's bitrate by 72 kbit/s.

Measurement results

The follwoing graphs show the effect on the waveform (top) and on artifact levels (bottom) of different codecs at different bitrates. The effect of the waveform is the RMS level of the difference between the compressed and the original signal, versus the original signal. The levels in dB's show how high the added noise level is, in absolute values. For MicroMPX these levels are flat across the spectrum, for MPX we have taken the average value.

Assumptions

MicroMPX is a codec intended to transmit MPX signals, not just any random data with high frequencies in it. All the measurements in this post were performed with a "valid" MPX signal (i.e. audio, pilot, stereo data and RDS).

If you try to encode signals of "invalid" MPX signals – e.g. a tone in the RDS area of the spectrum – the (meaningless) results will vary wildly.