Digital Audio Broadcasting
You may legitimately be wondering what DAB has to do with Musaeus. The answer is that Musaeus’ founder, Richard Black, is a hi-fi reviewer (principally for Hi-Fi Choice) under another of his hats and was one of the first reviewers in the UK to evaluate DAB tuners. His interest in the subject stems from those reviews, and from the somewhat disappointing results he obtained from the DAB system in general.
A little background. DAB (also known as Digital Radio) is touted as the successor to FM for high quality radio broadcast applications in Europe. It is a digital audio system, which uses MPEG2 data reduction to achieve data rates between 48kbit/s (very lowest end speech stations) and 256kbit/s (specified maximum) for each radio station. Claimed advantages are:
1. ‘Near CD quality’ sound (or according to some rather less scrupulous publicists, ‘CD quality’);
2. Absence of multipath distortion, hiss, fading and most other radio artefacts;
3. No need to retune car radios as the listener drives around since all repeaters for a given station are on the same frequency (this sounds impossible but works thanks to some very clever radio and digital audio engineering);
4. Lower total radiated power required to cover a whole country, due the additive effect of nearby transmitters as mentioned above;
5. More stations (national and regional) can be broadcast in the same bandwidth;
6. Heavy studio compression is not required since compression metadata can be sent with the audio bitstream and used to control (at the listener’s option) the dynamic range in the receiver.
By and large, these claims are met. Of course lack of interference is the usual deal with digital: when it starts to fail, performance ‘falls off a cliff’ very fast, but so far experience with DAB in properly covered areas (e.g. London) suggests that a piece of wire hung more or less any old how will indeed pick up interference-free radio.
The problem with sound quality is not that it is grossly or obviously distorted, but that, as so often seems to happen with data-reduced systems, it fails very subtly to meet one’s expectations. At best, sound is very close to CD, but at worst it develops a very high-frequency ‘warbling’, added to the music, which is not always noticeable at first but it one of those things that one spots quicker with experience. This warbling is sometimes referred to by the amusing name of ‘space monkeys’.
The reasons for that become clear when one examines a spectrum analyser display of DAB material. The first thing that one learns very quickly is that despite the apparent audio bandwidth of DAB being 24kHz (it uses a sampling rate of 48kHz), the actual broadcast bandwidth is at best 17kHz and frequently more like 15kHz - exactly the same as FM. Why? Because MPEG2 is a ‘perceptual coder’ which discards information on the basis of ‘least audible first’ and most of the time it regards information above 17kHz as inaudible - indeed this is provably a pretty good assumption in many cases. Certainly, the first impression most listeners have when listening to DAB is not of dullness, which rather proves the point.
More serious is that MPEG2 decides on a moment-by-moment basis how much audio bandwidth to allow through and with bright sounds the amount of treble tends to vary very fast, leading to the ‘warbling’.
Proposals for improvement
Alan Tutton of the BBC was quoted in New Scientist early in 2000 as suggesting that one possible improvement to DAB might come from using a sampling rate of 32kHz rather than 48kHz, thus limiting the bandwdith to a (generally satisfactory) 16kHz before encoding. However, my experiments suggest that there is little or no benefit in this: indeed, MP3 (a similar coder to MPEG2 in many ways, and one which is freely available to experiment with as well as very flexible in handling of sampling and data rates) seems invariably to give the best results at the highest sampling rates. What does yields a surprising degree of improvement in DAB sound, even at the receiving end, is a simple bandwidth limitation, effectively filtering off the varying and disturbing treble.
Most of my experiments have been carried out on material from BBC Radio 3 (note for non-UK readers: the BBC’s national classical-music station) since that is what interests me most and the station has some of the best studio practice and sound quality generally. It also uses a high data rate of 192kbit/s (the rest use 160 or 128kbit/s for music and less for speech). I have found that a bandwidth limit at 14.2kHz seems to give the best compromise, cutting out the warbling almost entirely without significantly affecting perceived brightness on most music (very bright sounds such as cymbal sound a little bit rough, but then they do before the treatment anyway and that kind of effect is arguably a lot easier to live with and ignore).
A recent broadcast provided a perfect example of the warbling effect and what can be done to mitigate it. If you can put up with the boredom of the rather slow download (in the circumstances there would obviously be no sense in putting anything other than uncompressed PCM audio on the web site: as a result the file is approx. 1MB for 6 seconds of audio), I strongly recommend that you download this file and listen carefully to the same chord, first exactly as captured from a DAB tuner and then after low-pass filtering with an extremely sharp FIR filter at 14.2kHz. The chord in question is the final chord of Alan Bush’s piano concerto. Note that computer loudspeakers will probably tell you very little about this! [Note: shortly after this page was published, a fellow audio engineer wrote to me with the news that, on the contrary, the effect was clearly audible on a pair of $100 computer speakers.] If possible, listen over good quality headphones. Alternatively, burn the file to a CD (they only cost 50p, after all) and play it on your hi-fi. In order to make that possible, I have sample-rate-converted the file from 48kHz to 44kHz and you’ll just have to take my word that that has very little effect on the sound quality. If you are curious as to what FM sounds like by comparison, try this file - the same chord from the same broadcast, recorded from a Revox A76 analogue FM tuner (one of the best ever made) fed by a rooftop antenna and recorded on a Marantz CD recorder. I’m interested in your reactions.
Now there are various implications one can draw from this. It is encouraging that so simple a fix can have such a useful effect, and although the filtering in the example given here was done offline (using Cool Edit) there is absolutely no reason why someone couldn’t make a simple inline box, digital in and digital out, that could sit between a DAB tuner and a DAC and provide just exactly such filtering. In fact I wish someone would make such a device, since with a change only in cuttoff frequency it could also provide a2i filtering between CD player and DAC. If you are a manufacturer interested in producing a gadget like that, please get in touch! I would love to discuss it with you - I simply haven’t time to learn DSP programming and hardware and would only ask a very modest donation to charity in lieu of a consultancy fee! I could outline the concept to you in a single email.
However, it would surely be preferable to have the filtering performed before MPEG2 encoding, which would also free up more bits (in principle, anyway) for coding lower frequency sounds more precisely. But would broadcasters want to admit, even to themselves, that they were only broadcasting a narrower bandwidth on DAB than on FM? Probably not. However, there is a cleverer possibility which I hope someone will look at, which is to apply a little more processing power within the MPEG2 coder and ensure that treble extension cannot vary so fast. This is not trivial: it would be easy enough to do if the coder were only ever required to work offline, but of course a lot of broadcasting is still live so coding must necessarily be done in real time, with limits thus placed on the amount of delay - already high at nearly 1 second.
I have tried some experiments on similarly bandlimiting pop-music stations, including some transmitting at 128, 160 and 192kbit/s. At the highest data rate a bandwidth of about 14.5kHz seems optimal, while at 128kbit/s about 13kHz - 13.5kHz seems to be about as good as it gets. However, apart from the fact that evaluation of such stations is difficult because they frequently use some degree of signal processing (including compression: just because it’s not strictly necessary on DAB does not mean that they will not use it, unfortunately), some of them may be playing from MP3 jukeboxes or similar (I recently heard from a colleague that one radio station found its DAB broadcasts will on occasion have gone through as many as FIFTEEN separate codecs before reaching the listener - arrrrgh!), they may very likely be playing radio remixes of many songs - and sadly all too many current pop songs are horrendously badly produced in the first place, so the fact that they sound rough is not necessarily any reflection on the DAB system. But as far as I can tell, A) the principle of bandlimiting the audio does yield a (small) improvement and B) 128kbit/s is not by any means ideal for clean-sounding results with music. 160kbit/s is a bit marginal and I just wish that the broadcasters were using at least 192kbit/s for all ‘serious’ radio stations and 192kbit/s for any music station.
For further information about MPEG2 and data reduction algorithms in general, you could do a lot worse than look at the web site of the Fraunhofer Institute, one of the main developers of such systems, or look at the page of the world DAB forum.
Go to the Musaeus Technology page or the a2i page and read about some of our other little innovations.