Introduction to Wi-Fi radios

Complaint to BBC about the bit rate reduction on the on-demand Internet radio streams

Below is a copy of the complaint I sent to the BBC about their decision to reduce the bit rate levels of the on-demand Internet radio streams. I sent the complaint to the BBC a few days before the new AAC/AAC+ on-demand streams launched (which is why the launch of the AAC+ streams is referred to in the future tense in the complaint) in the hope that they would see sense and choose not to reduce the bit rate levels. However, in its infinite wisdom the BBC chose to ignore my suggestions and reduced the bit rates anyway, so the BBC has chosen to reduce the audio quality relative to what it would have been if they'd have left the bit rates unchanged.

Incidentally, someone from the BBC has claimed elsewhere on the Internet that I've sent "numerous complaints" about this, which is incorrect, because this is the only formal complaint I've ever sent to the BBC about the Internet radio streams.

Summary of complaint

I would like to complain about the decision to reduce the bit rate levels of the on-demand radio streams when they switch from using MP3 to the AAC/AAC+ audio format.

The main justification that the BBC has put forward for why the bit rate levels should be reduced is that higher bit rate streams suffer from reliability problems. However, evidence strongly suggests that the reliability of the on-demand streams is fine at the bit rate levels they're currently using. Therefore, reducing the bit rate levels simply amounts to reducing the level of audio quality relative to what it would be if the bit rates were not reduced.

By reducing the bit rate levels to 96 kbps AAC+, the BBC would be reducing the audio quality to well below that provided by 128 kbps WMA, which around a third of all UK commercial radio stations use for their Internet streams. There are also over 8,000 Internet radio stations that are using 128 kbps streams or higher bit rate levels.

There are also compelling audio engineering reasons for why 128 kbps should be used for Internet radio streams, and the poor-sounding 96 kbps WMA live streams that the BBC launched a couple of months ago are a shining example of why 128 kbps should be used for Internet radio streams that carry stereo music, and not 96 kbps.

Because of the huge amount of bandwidth that the iPlayer TV streams are consuming, and the amount of money that the BBC spends transmitting DAB each year, not to mention the bandwidth the iPlayer TV streams are expected to consume in the near future and the £40m - £70m per annum that providing universal DAB coverage is estimated to cost, I find it extremely difficult to believe that the BBC needs to reduce the bit rate levels of the on-demand radio streams in order to save bandwidth or money.

The BBC is also planning on launching HDTV streams on the iPlayer this year. Such streams would use a bit rate level that is 30 - 40 times that of the radio streams, so the BBC would be practising double standards if it chose to limit the audio quality of the on-demand radio streams by reducing their bit rate levels. Combining this with the fact that there doesn't look to be any legitimate reason for why the bit rate levels need to be reduced, the only logical explanation I can think of is that the BBC plans to deliberately limit the audio quality because of its long-standing bias towards DAB and its bias against Internet radio.

I have previously suggested that the reliability of the on-demand streams can be improved by using a large buffer size, but my measurements show that the buffer size has not been increased. I would therefore reiterate that if you have any concerns about the reliability of the on-demand streams then you should increase the buffer size used. The buffer size can be set arbitrarily high for the on-demand streams, which would make the streams arbitrarily robust for everyone other than people suffering from major problems, in which case the streams would rebuffer no matter what bit rate were being used. There is especially no justification for reducing the bit rate levels without even bothering to see what the effect of using a large buffer size is on the buffering statistics.

Full complaint

My complaint relates to the intention to reduce the bit rate levels of the on-demand radio streams when the streams switch from using MP3 to AAC/AAC+.

The bit rate levels used at present with MP3 are as follows: Radio 3 uses 192 kbps, the other national stereo stations use 128 kbps, and the mono speech stations use 80 kbps. I believe that the 128 kbps MP3 streams are going to be reduced to 96 kbps AAC+, and Radio 3's streams will be reduced from 192 kbps MP3 to 128 kbps AAC.

I believe that the BBC's justification for reducing the bit rate levels is that the AAC/AAC+ audio codec is more efficient than MP3, so that the reduction in bit rate levels would not lead to a reduction in the level of audio quality.

However, if the bit rate levels were not reduced, the audio quality would be signfiicantly higher than it is now as a result of changing from MP3 to AAC. Therefore, unless there are any other factors that necessitate a reduction of the bit rate levels, doing so actually amounts to deliberately reducing the level of audio quality relative to what it would be if the bit rate levels were not reduced.

The only factors that I can think of that might influence any decision to reduce the bit rate levels are as follows, which are discussed in-turn below:

• Reliability of streams
• Bit rates & audio quality
• Bandwidth
• Distribution costs
• BBC's bias towards DAB and against Internet radio

Reliability of streams

The main justification the BBC has given up to now for why the Internet radio streams should be distributed at lower bit rate levels is that there is apparently some concern associated with the reliability of higher bit rate streams, and it has been suggested that lower bit rate streams would be more reliable. However, evidence strongly suggests that there isn't a problem with the reliability of the on-demand streams at the bit rate levels they're currently using:

1. As the bit rate levels of the on-demand radio streams have remained unchanged since they were launched in June, I fail to see how there can be a significant problem with their reliability because, by the BBC's own logic, if there were a problem then surely the bit rate levels would have been reduced by now in order to improve their reliability. This has not happened, so the only logical conclusion is that there cannot be a signficant problem with reliability as things currently stand.
2. On 29th January, I measured that the buffer for the on-demand streams held a minimum of approximately 10 seconds' worth of audio. I then suggested that, if there is a concern about the reliability of the streams, the buffer size should be increased, because the larger the buffer size is the lower the probability will be that the buffer will empty. I measured the buffer size being used yesterday, and the buffer size has remained unchanged. If there were a problem with the reliability of the on-demand streams, I would have expected that the buffer size would have been increased to improve the streams' reliability. Again, this suggests that there isn't a problem with reliability at the current bit rate levels.
3. The view that 128 kbps streams are significantly less reliable than 96 kbps streams is not borne out in practice. If this theory were correct, you would logically expect that the 500 kbps and especially the 800 kbps iPlayer TV streams would be exceptionally unreliable. But that is not the case. And if 128 kbps streams were so unreliable, why do so many other radio stations use this bit rate level, whereas very few use 96 kbps?

Dual-threshold buffering on Flash

The iPlayer TV and radio streams use 'dual-threshold' buffering, which is described here:

Dual-threshold buffering allows you to use very large "upper threshold" buffer sizes without having any impact on the time it takes for streams to start playing -- that is determined by the "lower threshold" buffer size, and the upper and lower thresholds are set independently of one another.

You can therefore freely select how much audio you want to store in the buffer -- even storing up to 1 - 2 or more minutes' worth of audio if so desired. The larger the buffer size is the lower the probability will be of the buffer emptying. A buffer only holding 10 seconds' worth of audio is far too small for the on-demand streams, because the buffer size can be set arbitrarily high.

I'm very confident that if you employed a large buffer you would not have a problem with the reliablity of the on-demand streams at the current bit rate levels. The only users who would still experience buffering would be those who're experiencing serious problems, such as serious congestion, in which case it wouldn't matter what bit rate you used, the streams would still buffer anyway.

Bit rates & audio quality

A third of all UK commercial radio stations that also transmit on FM, AM or DAB (95 stations altogether) are now using 128 kbps for their Internet radio streams, and most of these stations have been using 128 kbps streams for the last 2 years. The vast majority of these streams are using 128 kbps WMA, which provides higher audio quality than the 96 kbps AAC+ that the BBC is planning to use for its stereo streams (other than Radio 3). This means that the BBC is deliberately choosing to provide lower audio quality than commercial radio provides. I fail to see how this is in the best interests of licence-fee payers.

There are also over 8,000 Internet radio streams available on the shoutcast.com Internet radio portal that are using bit rates of 128 kbps or higher.

There are very strong audio engineering justifications for why Internet radio streams should use 128 kbps:

1. In a blind listening test, 128 kbps AAC scored 4.52 out of 5.0, whereas 96 kbps AAC only scored 3.82 out of 5.0, where 5.0 is classified as being "CD quality". Therefore, reducing the bit rate levels from 128 kbps to 96 kbps with the AAC codec would result in a marked reduction in audio quality. Furthermore, the BBC is planning on using AAC+ rather than AAC, but the Nero AAC/AAC+ encoder uses a bit rate threshold of 85 kbps, above which AAC is used, because the Nero software developers will have determined through their listening tests that AAC provides higher audio quality at bit rates above that threshold, and AAC+ is used at bit rates below 85 kbps because AAC+ is better at lower bit rates. Therefore, the BBC is actually using the wrong AAC profile for the Internet radio streams, and that will increase the drop in audio quality relative to the quality provided by 128 kbps AAC, which is what would be provided if the bit rate levels were not reduced.
2. Audio is inherently variable in terms of how difficult it is to be encoded using perceptual audio codecs, such as MP3, WMA, AAC/AAC+ etc, and the level of difficulty varies over short timescales. Audio codecs use two different coding "modes" to encode the audio: variable bit rate coding (VBR) and constant bit rate coding (CBR).
   With VBR, when the audio becomes difficult to encode the encoder allocates more bits to encode the audio with in order to ensure that the audio quality isn't reduced. Unfortunately, however, Internet radio streams use CBR, and when the audio becomes difficult to encode with CBR the audio quality descends rapidly if the bit rate level is too low -- because the encoder simply runs out of bits, so it's unable to maintain a constant level of quality. It is therefore necessary to use higher bit rate levels with CBR in order to provide some 'headroom' to avoid the audio quality being degraded when it is difficult to encode.
   Obviously, there are significantly more bits available to the encoder when 128 kbps is being used comapred with 96 kbps streams, so 128 kbps streams are far less likely to suffer from such large reductions in audio quality happening, and any reduction in audio quality would always be far less severe with 128 kbps anyway.
3. A lot of the audio played on radio stations has been "transcoded", which means that the audio has been encoded in a lossy audio format prior to the final lossy encoding that produces the streams that listeners receive. BBC R&D wrote a research paper that covers this topic well. It is widely known that transcoding, or cascaded audio coding as the BBC R&D paper calls it, leads to a significant reduction in audio quality, as the following sentence from the paper sums up:
   
   "the reality is that the cumulative effects of bitrate reduction throughout the broadcast chain is far from negligible."
   
   And that's putting it very mildly. The graph showing the results on page 9 of the research paper also show a very strong correlation between using low "emission" bit rate levels (i.e. the bit rate levels of the streams that listeners receive) and the audio quality being poor. Therefore, it is necessary to use higher bit rate levels in order to compensate for the use of transcoding, or else the quality will be poor.

Taking both of the above points into consideration, it is hardly surprising that the BBC's recently-launched 96 kbps WMA streams sound as poor as they do -- and these 96 kbps WMA streams are using the very same "Coyopa" architecture that the on-demand streams under discussion will be using. Furthermore, the WMA and AAC/AAC+ audio codecs perform very similarly in terms of audio quality in the 96 - 128 kbps bit rate region, so the 96 kbps AAC+ streams are likely to sound poor also.

96 kbps AAC+ may have proven to be adequate for use on the higher quality iPlayer TV streams, but there are large differences between audio on TV and that on radio: Audio on TV primarily consists of speech, whereas music is far more difficult to encode than speech is, so music requires the use of significantly higher bit rate levels.

Audio that's broadcast on the radio is also far more likely to have suffered from poor audio engineering, such as being transcoded, than audio on TV.

Overall, a bit rate of 96 kbps is insufficient for the Internet radio streams, and 128 kbps is far better suited for this purpose -- because of the issues with using CBR and transcoding, a higher bit rate than 128 kbps should ideally be used to compensate for these problems.

Bandwidth

The combined bandwidth consumed by the live and on-demand Internet radio streams in March 2008 was 461 TB/month.

In comparison, the bandwidth consumed by the iPlayer TV streams had already reached 1,756 TB/month by May 2008, and quoting from an EBU Technical Review article documenting a series of phone interviews with Anthony Rose that took place in August 2008:

"We do something like 100 TB per day of streaming traffic."

In other words, the iPlayer TV streams were consuming around 3,000 TB/month, so the bandwidth for the iPlayer TV streams looks to have almost doubled between May and August -- presumably largely because the higher quality 800 kbps H.264 iPlayer TV streams were launched in that period.

There are also further quality and therefore bandwidth increases planned for the iPlayer TV streams for 2009, such as "in three months time we might be offering 1.5 Mbit/s", and HD streams are planned to be launched this year, which would use a bit rate of around 4 Mbps -- which is around 30 - 40 times higher than the bit rates used for the Internet radio streams.

Plans to allow consumers to watch iPlayer TV streams on their TV sets also looks set to increase the bandwidth required dramatically:

"But imagine in a year or two when we have a TV set-top box with an integrated iPlayer and millions of people using it, and each of them consuming 1.6 Mbit/s for a TV stream. The bandwidth required would be 10 times what it is now."

That would imply that the iPlayer TV streams would be consuming 30,000 TB/month, which is 65 times higher than the combined bandwidth that the live and on-demand Internet radio streams consumed in March 2008.

It is extremely difficult to believe that the BBC needs to reduce the bit rate levels of the on-demand radio streams in order to save what would be a paltry amount of bandwidth compared to that required for the iPlayer TV streams.

Distribution costs

Bandwidth costs account for the large majority of Internet streams' distribution costs, so the same argument applies to distribution costs as applied to the need to save bandwidth. However, I think it is still worth pointing out that bandwidth costs are falling rapidly, as the following quotes from the EBU interview with Anthony Rose show:

"[over the last two years] the price of bandwidth has declined dramatically"

"The cost of bandwidth is falling very rapidly and there is a lot of competition between the CDNs."

The cost of bandwidth is inversely proportional to the speed of Internet routers, whose speed has followed Moore's Law since the 1980s, as the following figure shows:

What this means is that the cost of bandwidth halves about every 18 months. So in the twelve months since the BBC Internet blog was published announcing the forthcoming quality improvements for the Internet radio streams, the cost of bandwidth should theoretically have fallen by 37%. Therefore, any decision taken 12 months ago to reduce bit rate levels is already out-of-date, and that decision should be reviewed in light of the large reduction in bandwidth costs over the past year.

Furthermore, the BBC is mid-way through the transition from a £6 million to an £11 million per annum contract for the transmission costs for the BBC's national DAB multiplex. So the BBC is probably currently spending around £8.5 million per annum transmitting its national DAB multiplex, and it is spending a further £3.6 million on DAB transmission costs for its local stations as well, making a total of £12.1 million per annum. In addition, it is estimated that it will cost between £40 million to £70 million per annum to distribute the BBC's national DAB multiplex once it has been extended to reach "universal" population coverage levels.

Considering the hundreds of millions of pounds that will be spent on transmitting the BBC's national DAB multiplex at universal coverage levels, and the exponentially increasing distribution costs for the iPlayer TV streams, it would be an insult to Internet radio listeners if the BBC claimed that it needed to save a relatively paltry amount by reducing the bit rate levels of the on-demand radio streams.

The BBC's bias towards DAB and its bias against Internet radio

BBC TV advertising for digital radio

The BBC has broadcast 21 high-impact TV advertising campaigns for DAB since 2002. In comparison, the BBC has broadcast zero TV advertising campaigns for Internet radio. Yet, according to the BBC's "values", the BBC considers itself to be:

"independent, impartial and honest"

I'm at a failure to comprehend how broadcasting 21 high-impact TV advertising campaigns for DAB and none for Interent radio could be classified as being "impartial" by any stretch of the imagination.

Admission that the BBC would prefer people to listen via DAB

When Simon Nelson was the Controller in charge of digital radio, he candidly admitted on Radio 4 Feedback that:

"Of course the BBC would prefer it if everybody listened to digital radio via DAB"

I've been told that this is no longer BBC policy, but the BBC supports the DRWG's recommendations, so it obviously still is BBC policy, because the DRWG's recommendations to government were as biased towards DAB as it is possible to get.

BBC's support for pro-DAB & anti-Internet radio recommendations for government

The BBC supports the recommendations made in the Digital Radio Working Group's (DRWG) final report, and Mark Friend had the most influential position within the DRWG as the chairman of the Technology sub-group, whose task it was to supposedly investigate and recommend which digital radio platforms should be used and which should not. In reality, it became abundantly clear that DAB had been chosen from the outset and that no other platforms were seriously considered. The DRWG duly recommeded that DAB should become the main digital radio platform, and Internet radio was not mentioned in any of the recommendations in the final report. So, based on the BBC's past history, I would expect that the BBC will only promote DAB from now until FM switch-off, and the BBC will withhold information about the benefits that Internet radio provides from the general public. The BBC therefore chose the protectionist path in full knowledge that this was completely against the best interests of licence-fee payers.

Diabolical audio quality of BBC Internet radio streams from 2003 - 2007

The audio quality of the BBC's Internet radio streams was diabolical over the period from 2003 to 2007. The following recordings of Radio 1's listen again streams in October 2006 are typical of the unlistenably bad quality provided over the entire period.

Recording #1 of Radio 1 on the BBC Radio Player 26th October 2006 (FLAC lossless version)

Recording #1 of Radio 1 on the BBC Radio Player 26th October 2006 (320 kbps MP3 version)

Recording #2 of Radio 1 on the BBC Radio Player 26th October 2006 (FLAC lossless version)

Recording #2 of Radio 1 on the BBC Radio Player 26th October 2006 (320 kbps MP3 version)

I measured the bit rates being used on numerous occasions, and the streams for Radios 1, 2 and 4 (the most listened-to stations) always used a bit rate of 32 kbps when I measured them, and Radio 3 and some of the other stations were using 40 kbps. This begs the question why the BBC was still using bit rates as low as 32 kbps when GCap Media had begun using 128 kbps WMA streams for all of its stations (around 55 stations in total) in early 2007, and literally thousands of Internet radio stations had been using 128 kbps MP3 for a few years by then as well.

I've recently read that the requirements for the new Coyopa encoders were drawn up four years ago, so you have to ask why it has taken four years for these encoders to be deployed? RealNetworks included the AAC+ audio codec in Real Player in January 2004, and AAC had already been available before that date, so why are the Real Player streams still using the Real G2 audio codec today when AAC and AAC+ are far superior codecs -- AAC+ is especially far and away the best codec to use at the very low bit rate levels that the BBC was using over that period. Why has the audio been transcoded at Maidenhead for the last 6 years when a leased line could have been installed for a few thousand pounds per annum? Even without a leased line being installed, why was the audio for the on-demand streams transcoded at all when the audio could have been automatically recorded and chopped up in London and sent to Maidenhead as files over the Internet at zero cost? Crucially, would any of these things have happened if this were DAB's audio quality we were talking about? Not a chance.

BBC delayed quality improvements for live streams in 2008 for no apparent reason

The BBC's live Internet radio streams were originally planned to start using higher bit rates to improve their audio quality in June last year, straight after the on-demand streams began using MP3 at higher bit rates. So why has the audio quality of the live streams not improved over the last 7 months (I'm assuming that the vast majority of people do have Real Player installed, so they would be served with Real Player live streams)? For example, the Real G2 codec looks to be able to use bit rate levels up to 96 kbps. So even without having to launch new audio formats for the live streams it looks to have been possible to improve their audio quality simply by turning up the bit rate levels of the existing Real Player live streams.

As this was such a simple solution, why hasn't this been done at any point over the last 7 months? Would this have happened on DAB?

BBC specified that DAB+IP radio prototype should only receive DAB broadcasts

In 2007, the BBC paid the Schulze and Webb design consultancy to design and build a DAB+IP radio prototype -- referred to as the 'Olinda project' -- that included Wi-Fi to provide Internet connectivity. However, the BBC specified to the design consultancy that the radio should only be allowed to receive DAB broadcasts Although this may only have been a prototype, this hardly instills a huge amount of confidence in the BBC's impartiality on this subject.

BBC intended to provide live streams at lower quality than the on-demand streams

Until late last year, the BBC was planning on providing the live Internet radio streams at lower audio quality than the on-demand radio streams. The BBC attempted to justify this by saying that higher bit rate streams suffer from reliability problems. The BBC also attempted to justify this by saying that live radio is also available via DAB, FM and on the digital TV platforms. But if you remove the reliability issue from the equation (because technologies are available that solve the reliability issue, such as automatic bandwidth detection and switching between streams at different bit rate levels), that only leaves the fact that the BBC considered that it was acceptable to degrade the audio quality of the live Internet radio streams because live radio is also available on other platforms. By definition, that is being biased against the live Internet radio streams.

etc, etc

Conclusion

As the BBC is obviously biased towards DAB and biased against Internet radio, and as there doesn't seem to be any legitimate reason for reducing the bit rate levels of the on-demand Internet radio streams, the only logical explanation is that the BBC wants to reduce the bit rate levels in order to limit the audio quality they provide, and that is because the quality would otherwise be a lot better than that provided on DAB.
 
 

Comments

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