DAB vs DAB+ technology

The saddest thing about DAB adopting the AAC+ codec is how long it took them to upgrade the old DAB system. The DAB system was designed in the late 1980s, and by 1990 all of the main technologies had been chosen, and all of these main technologies remained unchanged until they announced the upgrade to the system at the end of last month.

MP2 to AAC+

In order to choose which audio codec to use on the original DAB system, the designers conducted a listening test at Swedish Radio in 1990 with a number of prospective audio codecs to see which was best. The two 'winners' were the MP2 and MP3 audio codecs. A long debate apparently followed about which codec to go for, but unfortunately the designers made the diabolical decision of choosing to use MP2 rather than MP3.

They chose to use MP2 due to the lower computational complexity of MP2 decoders, but using MP3 would have made only a tiny difference to the overall power consumption of a receiver, and yet it would have drastically improved the DAB system.

Then, in 1993, following research that showed that performance could improve at low bit rates by removing the requirement to be backwardly compatible with MP2 and MP3, work began on the AAC (Advanced Audio Coding) codec. AAC was eventually standardised in 1997.

So, the DAB system designers could have chosen to move to using AAC at any point from 1993 onwards, but instead they stuck their heads in the sand.

The audio codec is the most important single technology used in a digital radio system, because it directly affects the number of stations that can be transmitted in the available spectrum, the audio quality that the stations can be provided at, and the cost of transmitting the stations. And the fact that it has taken them 16 years to replace the totally inadequate MP2 audio codec is, in my opinion, pretty incompetent considering that they had about a decade to realise that MP2 is so incredibly badly suited to the requirements of a digital radio system.

I can remember when the BBC slashed the bit rates of their radio stations on DAB in December 2001; I'd owned a Psion Wavefinder DAB receiver for a couple of months and was totally unaware of the BBC's plans to slash their bit rates in order to fit on the 5 new stations they planned to launch. But the change in audio quality was dramatic: Radio 1 and Radio 2 went from sounding good to sounding diabolical and it was clear that something was massively wrong. So, as I'd studied DSP and digital communications at university I decided to buy a book about digital audio (Principles of Digital Audio, by Ken Pohlmann — highly recommended if you're interested in the subject, and it includes an excellent 60 or so page chapter on perceptual audio coding, i.e. audio compression) to investigate what was wrong. The answer was immediate in coming as soon as I turned to the chapter on audio compression: MP2 was the totally wrong audio codec to use at bit rate levels of 128 kbps, which the BBC had just reduced their most-listened-to music stations to, and which many of the commercial radio stations were also using. 

Audio codecs have a 'sweet spot' where they're optimised to perform well at: MP2 is meant to be used at high bit rates, such as 192 kbps and above; MP3 is optimised for lower bit rates levels of around 128 kbps, and AAC was optimised to be used at 96 kbps. And DAB's new audio codec, AAC+, is optimised for very low bit rate levels of around  64 kbps.

To this day I simply have no idea whatsoever how so many supposedly intelligent and experienced engineers working on DAB could go so many years without spotting the glaringly obvious problem that the MP2 audio codec was totally unsuited for use on a digital radio system due to it needing to be used at high bit rates yet DAB only had a limited amount of spectrum available for it (the UK currently uses 12 MHz of spectrum for DAB — in comparison, analogue TV consumes around 400 MHz of spectrum). High bit rates also mean high transmission costs — £200k per annum to transmit on a local multiplex at the kind of audio quality level that was originally envisaged on DAB — and for a radio industry where, according to Ofcom, 40% of all stations either make no profit at all or indeed trade at a loss, there was only going to be one outcome with DAB using MP2: the audio quality would have to be low.

It may be a strong word to use, but I happen to think it is entirely appropriate on this occasion: the decision to use the old DAB system in the UK was simply grossly incompetent.

Single to dual-layer error correction coding

DAB's other main Achilles' heel is its weak error correction coding, which is also going to be upgraded in the new DAB+ standard. 

If you own DAB, then unless you're very fortunate you will have experienced the highly annoying 'bubbling mud' sound that accompanies poor reception quality. This bubbling mud sound that you hear is the consequence of too many errors going uncorrected by the error correction coding system. 

For example, I have poor reception quality on one of the 4 DAB multiplexes I can receive, and yet I live only 5 miles away from the transmitter for this multiplex. I don't have a problem with reception quality on FM stations that are transmitted from this multiplex, so although increasing the transmitter power of the DAB multiplex would help the situation, the error correction coding is obviously too weak when a multiplex transmitted from just 5 miles away has reception quality so bad that the whole multiplex is unlistenable (the technical reason why there is a problem with the error correction coding on DAB is that the error correction 'code rate' used to protect the MP2 scale factors is too high (a higher code rate means weaker protection, and vice versa) for the importance of the scale factors to the correct playback of the audio).

The problem with the original error correction coding used on DAB is that it consists of a single 'layer' of error correction coding rather than the more modern approach of using 2 or more layers (the layers are described as being "concatenated"), where the outer layer(s) of error correction coding is used to catch the errors that the inner layer fails to correct. So, with DAB, if the one and only layer of error correction fails to correct the errors, those errors are passed directly on to the MP2 audio decoder, and that's why we hear the bubbling mud when too many errors have failed to be corrected.

The new DAB standard will use a second outer layer of error correction coding, called Reed-Solomon (RS) coding, for the purpose of greatly improving the strength of the error correction. The RS coding adopted for the new DAB standard is the same as that used on the DVB-T (Freeview) system, where it reduces the bit error rate (BER) from around 2 x 10^-4 down to the 10^-11 required for video reception (video can tolerate far fewer errors than audio can tolerate). What this means is that by adding RS coding — and with all else staying the same, including transmitter powers —  basically everybody that currently receives a DAB multiplex but experiences bubbling mud or intermittent dropouts would have their reception quality transformed to being perfect if RS coding were used. 

And, similar to the situation with the audio codec, RS coding was already available in the late 1980s when they were designing the DAB system — RS coding is used to correct errors on CDs, so it's been around since the early 1980s — but they went with the less computationally complex but weaker error correction coding scheme that has proved to also be inadequate for the job in hand.

All in all, the original DAB system is a poorly designed system, and the sooner it is ditched the better for all concerned.

Comments

Bookmark with:
Digg	del.icio.us	Reddit	Facebook
Google	Stumbleupon	Slashdot