DAB vs DAB+ technology

Introduction

The DAB system was designed in the 1980s, and because the technologies it uses are so old -- the technologies are unchanged to this day -- DAB is a very inefficient system by today's standards. 

In 2003, new systems emerged that had been designed to enable mobile TV, such as DVB-H and T-DMB, and these systems could also carry digital radio. However, crucially, these systems used the AAC+ audio codec and Reed-Solomon error correction coding, and the combination of these two technologies made DVB-H six-times as efficient as DAB and T-DMB four-times as efficient as DAB.

Because of its inherent problems, broadcasters and governments from numerous countries became opposed to using the old DAB system, so WorldDAB (now called WorldDMB) was forced to upgrade the old DAB system or risk seeing the UK, Denmark and Norway stranded using the old DAB system while all other countries would have chosen a more modern system instead. The upgrade they came up with is called 'DAB+', and this page compares the technologies used on DAB with those used on the new DAB+ system.

1.1.1  MP2 Codec Technology

Codec Type

MP2 is what is called a 'sub-band codec', which means that the linear PCM input audio signal is first split by a filterbank (a poly-phase quadrature mirror filter (PQMF)) into 32 equal-bandwidth 'sub-bands'. The input signal is then analysed by the psychoacoustic model (a model of the human hearing system), which determines which of the 32 sub-bands will be perceptible by humans and which won't be. Those that are deemed to be imperceptible are not transmitted, whereas the subbands that are deemed to be perceptible are all encoded.

As the input signal is simply split into 32 subbands and the signal is not transformed into the frequency domain, MP2 compression takes place in the time domain, as is always the case with subband codecs.

Frequency Resolution

Referring to Nyquist's Sampling Theorem:

F_s >= 2 B

where F_s is the sampling frequency and B is the bandwidth of the signal. For a sampling frequency of 48 kHz, which the DAB system uses, MP2's frequency resolution (the bandwidth of each subband) is:

MP2 frequency resolution = (48 kHz / 2) / 32

MP2 frequency resolution = 750 Hz

Stereo Coding

MP2 can encode stereo signals either using discrete stereo coding or joint stereo coding.

Discrete stereo coding consists simply of encoding the left and right channels separately. Discrete stereo coding is only suitable for high bit rates such as 192 kbps and above.

Intensity joint stereo coding consists of encoding the left and right channels together above a certain frequency threshhold that is set on a frame-by-frame basis by the encoder. Intensity stereo is the only type of joint stereo that MP2 can use (other codecs can use the superior mid/side joint stereo method) and intensity stereo can be used from a frequency of 2.25 kHz and upwards -- i.e. it can be used over the top 90% of the audio band, and on 128 kbps MP2 services on DAB it tends to be used over the whole of the available range, thus 128 kbps radio stations are actually panned mono rather than stereo.

For each subband, the amplitudes of the left and right channels are added together then encoded along with a vector for each subband that equates to the ratio between the average energy (the intensity) of the left and right channels. Joint stereo is invariably used on MP2 at bit rates of 160 kbps and below, because at such low bit rates it is preferable to accept some degradation of the stereo image in order to save some bits so that more bits can be used to more accurately represent the audio samples.

Intensity stereo is a lossy coding method, because the relative phase information between the left and right channels is lost when the left and right channel samples are added together. 

The ear uses the relative time-difference between sounds in the left and right ears to determine the direction from which the sound came, and it is therefore this relative phase information that creates the stereo image. Because the relative phase information above a certain frequency threshold is destroyed, this is why the 128 kbps DAB radio stations either have a very poor stereo image or the stereo image totally collapses.

The encoder can choose on an audio frame-by-audio frame basis whether to use joint stereo or discrete stereo coding. At bit rates as low as 128 kbps the encoder forces joint stereo for all frames, and at 160 kbps either all or virtually all frames use joint stereo coding. Although most radio stations that use 192 kbps -- for example on digital satellite -- use discrete stereo coding (which disallows the use of joint stereo for any audio frames), it is still better to use joint stereo at a bit rate of 192 kbps so that the encoder can have the option to choose joint stereo coding for audio frames where it would be beneficial. Unfortunately, the BBC misguidedly changed Radios 1-4 on Freeview (which all use 192 kbps) from being joint stereo to discrete stereo in 2004, and the audio quality of these stations on Freeview has never been as good since.

Sweet Spot

Karlheinz Brandenburg, the inventor of MP3, describes the sweet spot of an audio codec as follows:

MP2's sweet spot is at 192 kbps.

Just for comparison, Karlheinz Brandenburg describes the sweet spots of MP3 and AAC as follows:

"For Layer-3 [MP3] this target bit-rate is around 1.33 bit/sample (i.e. 128 kbit/s for a stereo signal at 48 kHz), for AAC it is around 1 bit/sample (i.e. 96 kbit/s for a stereo signal at 48 kHz)."

1.1.2  AAC+ Codec Technology

Codec Type

AAC+ is the combination of the standard AAC (LC AAC -- low complexity AAC) codec with Coding Technologies' Spectral Band Replication (SBR) technology -- the AAC audio codec encodes the bottom half of the audio spectrum, and SBR encodes the top half of the audio spectrum.

AAC is a transform codec, which means that blocks of input audio samples are first transformed into the frequency domain by means of a modified discrete cosine transform (MDCT), and compression takes place in the frequency domain, not the time domain.

SBR is based on the fact that there is a strong correlation between the top and bottom halves of the audio spectrum due to the presence of harmonics of the lower frequencies. SBR works by transposing the bottom half of the audio spectrum to the top half, and then modifying this top half of the spectrum so that it resembles the actual top half of the audio spectrum more closely. The SBR data only consists of the modification information, and this only requires an ultra-low bit rate channel of between 1 - 3 kbps, which is far lower than the bit rate that would be required if the top half of the audio spectrum is encoded by any other audio coding method. This is why AAC+'s official name is HE AAC -- High Efficiency AAC.

Frequency Resolution

AAC's frequency resolution varies depending on the statistical properties of the input signal: typically the signal has stationary statistics and a 2,048-point MDCT is performed, which gives a frequency resolution of 23 Hz for a 48 kHz sampling frequency input signal. For transient signals the AAC encoder reverts to a 256-point MDCT in order to improve the time resolution to avoid pre-echo.

Stereo Coding

AAC can use the following stereo coding types:

Discrete stereo coding - as explained in the MP2 section above

Mid/side joint stereo coding consists of forming the sum of the left and right channels (L+R) and the difference between the left and right channels (L-R) and encoding these sum and differences separately:

M = L + R

S = L - R

The decoder then carries out the following equations to return the L and R signals:

Left = M + S = 0.5 x ((L+R) + (L-R)) = 0.5 x 2L = L

Right = M - S = 0.5 x ((L+R) - (L-R)) = 0.5 x 2R = R

Unlike intensity stereo coding, which is a lossy form of joint stereo coding, mid/side joint stereo coding is lossless, which means that the original information is returned without losing any of the original information. Therefore, mid/side joint stereo coding does not have the collapsing and non-existent stereo image problems that MP2 has at low bit rates.

Intensity stereo coding - as explained in the MP2 section above

AAC allows the encoder to choose between the stereo coding modes listed above in a more flexible fashion than MP2 allows.

Sweet Spot

AAC+'s sweet spot is at approximately 64 kbps.

MP2 vs AAC/AAC+ performance

To compare audio codecs, listening tests are carried out that follow the ITU BS.1116 standard so that objective comparisons can be made between audio codecs and between different bit rate levels. 

The BS.1116 standard defines that testers should grade the audio quality according to the the following impairment scale:

The following table shows the scores achieved in listening tests for MP2, AAC and AAC+:

These results are shown graphically below:

The performance of AAC/AAC+ is therefore vastly superior to that of MP2 both in terms of the absolute level of audio quality that can be achieved (at reasonable bit rate levels) and in particular in terms of the coding efficiency.

1.1  Why is AAC+ so much more efficient than MP2?

Put very simply, MP2 was not designed to be efficient, whereas AAC+ is the culmination of over a decade's worth of advances in audio coding since MP2 was chosen to be used on DAB, and AAC+ is designed to be as efficient as possible.

1.1.3  Comparison of MP2 & AAC+ Codecs

The most obvious and largest difference in efficiency between the two codecs is due to AAC+'s use of SBR, which only consumes a bit rate of between 1 to 3 kbps to encode the entire top half of the audio spectrum, compared to MP2 (and all other non-SBR codecs) that consume a relatively large (not 50%, but still a very sizeable percentage) of the overall bit rate on encoding the upper half of the spectrum.

An inherent major problem with MP2 is its 750 Hz frequency resolution, which stems from the fact that the input signal is only split into 32 subbands. What the frequency resolution determines is how finely redundancy can be removed -- which is the key to reduced bit rate audio coding. With MP2's 32 subbands, if there is a frequency component that is deemed to be perceptible in a subband then that whole subband must be encoded. In comparison, the very fine frequency resolution of the transform codecs -- AAC's frequency resolution is just 23 Hz for signals with stationary statistics -- allows just those frequency components that are perceptible to be encoded, and the rest discarded, which is far more efficient than the way MP2 works.

The effect of too many subbands having to be encoded because the psychoacoustic model deems there to be at least something perceptible in that subband is that the available bit rate is spread too thinly, so there is an insufficient number of bits available to encode the perceptible subbands, which leads to an increase in the quantisation noise (coding noise) level, and in turn audio artefacts become perceptible.

This can most readily be perceived on pop and especially rock music, which tend to have a wideband spectrum and the dynamic range has already been compressed when the CD was mastered, and the radio station simply flattens the audio spectrum further by its use of audio processing. This results in a large number of subbands being deemed to be perceptible, which requires them to be encoded, resulting in dreadful definition, the stereo image is non-existent and it degenerates into a ridiculously low quality wall of sound. 

Transform codecs, with their much finer frequency resolution and their ability to remove more redundancy unsurprisingly perform much better with these more challenging to encode types of music.

Finally, MP2 being limited to using intensity stereo is another major Achilles heal compared to AAC+ and all the other audio codecs that do allow mid/side joint stereo coding. As mentioned above, only at a bit rate of 192 kbps and above does the encoder really begin to choose when intensity stereo is actually beneficial rather than having it forced upon it due to their being insufficient bits to use discrete stereo even when the signal demands it. This isn't a problem with mid/side joint stereo, because it doesn't destroy any of the phase information that intensity stereo does.

Overall, MP2 should not really ever be used at bit rates below 192 kbps, and 128 kbps is simply far too low a bit rate to provide audio quality that should be expected on a modern digital radio system.

Error Correction Coding

Error correction coding is the "heart" of a wireless digital communication system, and without it applications such as digital terrestrial TV, digital radio and mobile TV wouldn't be feasible. The error correction coding scheme used on a digital radio system is important for the following two main reasons:

• it determines how robust reception will be, because reception problems occur when the error correction coding fails to correct a sufficient proportion of the bit errors that inevitably occur with transmission over the mobile channel
• it affects the spectral efficiency of the system, because a stronger error correction coding scheme can correct more errors than a weaker one, so stronger error correction coding schemes enable the capacity of a multiplex to increase 

An error correction coding scheme for a digital radio system must take into account the error performance of the audio bitstream it is protecting.

DAB's UEP error correction coding

DAB uses UEP convolutional error correction coding, where UEP stands for unequal error protection. 

Audio data is grouped into audio frames, and some parts of the audio frame are more sensitive to errors than other parts, and UEP protects more strongly the parts of the audio frame that are more sensitive to errors and vice versa.

The strength of a particular type of error correction coding can be varied by changing its "code rate", R_c, and a lower code rate will result in stronger protection and vice versa.

The figure below is adapted from the "Digital Audio Broadcasting: Principles & Applications" book, and it shows how DAB's UEP applies stronger error protection to the header, scale factors, PAD and so on, because these parts of the audio frame are important to the correct playback, and lower protection is applied to the sub-band samples (the actual encoded audio samples). The height of the blocks in the figure denote how strongly this part is protected.

Group 1 contains important information for things like synchronisation and audio stream information; group 2 contains the scale factors, which scale the subband samples (these are form the exponent of a crude floating point number system); group 3 contains the subband samples (these form the mantissa of a crude floating point number system to go with the scale factors); and group 4 consists of the PAD (programme associated data) and scale factors CRC (cylic redundancy check).

The R_c values quoted in the figure are those used for 128 kbps using Protection Level 3 (PL3), which is by far the most widely used Protection Level.

Although from the figure you might at first glance think that the main problem would be with the sub-band samples, because they have the have the weakest error protection, the main problem is the insufficient protection of the scale factors.

The scale factors form the exponent of the crude floating point system used to encode the subband audio samples, and any errors in these scale factors should be detected by the scale factors' CRC check. When such errors are detected this leads to either muting or crude error concealment techniques to be used for the affected subbands which produces the "bubbling mud" sound that accompanies poor DAB reception quality.

This problem with the error protection of the scale factors is that DAB's error correction coding scheme uses convolutional coding (which is not by any means a strong form of error correction when used on its own), and the code rate used to protect the scale factors is only 8/18, or 0.44. Only using a convolutional code at a code rate of 0.44 to protect something as crucial to the correct playback of digital audio as the scale factors are is far too weak, and it is unsurprising that reception problems are rife on DAB. 

The MP2 Robustness Myth

One thing that proponents of the old DAB system have consistently claimed is that MP2 is somehow more robust for use on digital radio systems than other audio codecs are. This view is typified in a comment made by Quentin Howard, the chief executive of Digital One and the current President of WorldDMB, when he said:

"... AAC+ and WM9 [are used] in other applications and an enhanced Reed-Solomon layer of error correction [is] available for these more fragile encoding algorithms."

The argument put forward by the proponents of the old DAB system goes as follows: Audio codecs such as MP3, AAC and AAC+ must use extra error correction coding to protect them whereas MP2 doesn't need any extra error correction coding to protect it on DAB, therefore MP2 must be more error-robust than the other audio codecs. 

This is simply completely false.

As discussed above, DAB uses UEP to protect MP2, and the reason it uses UEP is because both the length of an MP2 audio frame and the groups within each audio frame are fixed, so UEP can easily be applied. And it is only the use of UEP on DAB that makes it appear as though MP2 is more robust than other audio codecs, when in fact it is no more robust.

The length of audio frames for MP3, AAC, AAC+ etc is not fixed, therefore it is not as easy to use UEP with these other audio codecs -- although it is not impossible, because DRM uses UEP to protect AAC+.

The proponents of the old DAB system are simply failing to understand what I mentioned at the beginning of the section on Error Correction Coding, which is that it is better to use stronger error correction coding because this allows the capacity of a multiplex to increase. And indeed, DAB+ is using EEP (equal error protection) convolutional coding along with an outer layer of Reed-Solomon coding, which is far stronger than the error correction coding scheme used on DAB, and this will allow the multiplex capacity on DAB+ to increase by about 30-40% compared to the capacity of a multiplex using the old DAB system -- unless the broadcasters decide to greatly extend the coverage area rather than take advantage of the increase in capacity.

Demonstration of why MP2 is no more robust than other codecs

For the DAB proponents' claim to be true then MP2 must be more robust than other codecs when both audio codecs are using the same error correction coding scheme . 

So in order to demonstrate that they're wrong, I've written a program that simply adds bit errors to files at random. You can download an executable of the program I've written here, and the C++ program file is here. Here's three files you can test the program with:

128 kbps MP2 file with no errors added

128 kbps MP3 file with no errors added

64 kbps AAC+ file with no errors added

And here's the same files with errors added where the BER (bit error rate) is 10^-4:

Same 128 kbps MP2 file with errors added with a BER of 10^-4

Same 128 kbps MP3 file with errors added with a BER of 10^-4

Same 64 kbps AAC+ file with errors added with a BER of 10^-4

None of them are acceptable to listen to, but they're about as bad as each other, and I would say that the MP2 file is actually arguably worse than the MP3 and AAC+ files. 

To run the program, copy the above files with no errors added to them or some of your own files (see note below though) to the same directory in which you've put the aac_mp2_errors.exe program; run the program, enter the audio file's filename when requested, choose a bit error rate (BER) value (e.g. enter 1e-4), and say no to using RS coding (enter the letter n).

A suitable BER value is 10^-4, because this is the typical BER figure quoted for digital radio systems, such as DAB and DRM. A BER of 10^-4 means that there is one bit error every 10,000 bits.

Note: You might run into problems playing back AAC/AAC+ files that you've encoded yourself after you've added errors to them, because audio you encode at home isn't expected to have any errors, so things like Nero's AAC/AAC+ encoder must use different settings with respect to error detection and Winamp sometimes won't play it back, especially when the BER is high -- this is why I've used an MPEG-2 AAC+ file that use ADTS headers (which are similar to the headers used for MP2 and MP3) above rather than an MPEG-4 AAC+ file encoded using Nero.

Theory for the program

The above program adds bit errors at random to the audio files, and this simulates EEP (equal error protection) convolutional coding with an infinite length (i.e. ideal) time-interleaver. The program acts identically for the different audio formats, so it is a fair comparison.

The reason why MP2 is no more robust than the other codecs can be ascertained by looking at the figure of the MP2 audio frame above. All audio formats that are used on broadcasting systems or live Internet streams are split up into audio frames like MP2 is, and these audio frames consist of a small percentage of data that is very important to the correct playback of the audio, and if an error lands within this important part of the audio frame then there's a high, or at least higher, probability that the error will be perceivable to the listener than if the error lands in the less important part of the audio frame.

Looking at this mathematically: The audio frame header is the most important part of the audio frame, because errors that land in the audio frame header are the most likely to lead to an audible disturbance. The audio frame header accounts for approximately 6% of the length of an audio frame for a 128 kbps MP2 stream, so the probability that an error will land in the audio frame header is:

Probability of error landing in audio frame =  

So if you consider the case of MP2 and AAC being protected by identical EEP convolutional coding along with an interleaver of infinite depth (the job of an interleaver is to make the errors uniformly randomly distributed), the areas of the audio frame for both MP2 and AAC will have identical protection, so the BER (bit error rate -- which is the proportion of bits that are in error (it's not really a rate, but that's the name for it)) will be identical for both audio codecs and the 

One thing that proponents of the old DAB system have consistently claimed is that MP2 is somehow more robust than other audio codecs. For example, one classic quote made by the current President of WorldDAB, Quentin Howard, who is also the chief executive of the UK national commercial DAB multiplex operator, Digital One, is as follows:

"Spurious claims from some quarters that MPEG-1 Layer2 audio is outdated or inefficient is a failure to understand the beauty of the way the frame length of MPEG and COFDM co-exist and the benefit of UEP which together deliver a very robust audio experience. Eureka-147 allows for other audio coding, of course, with BSAC being used in Korea, AAC+ and WM9 in other applications and an enhanced Reed-Solomon layer of error correction available for these more fragile encoding algorithms."

I find it hard to put into words just how ridiculously contradictory I think that statement is, because on the one hand it recognises the UEP makes reception quality more robust, but it then goes on to completely ignore the benefit that UEP brings to MP2 and accuses the more modern codecs of being "fragile".

What seemingly all the DAB supporters get wrong is that it is ONLY the UEP coding that makes them think that MP2 is more robust than other codecs. Without UEP you have EEP  — Equal Error Protection, where the whole audio frame is protected with the same code rate, so all sections of the audio frame are protected with equal strength — and if you protected MP2 using EEP then it would be no more robust to errors than the more modern audio codecs.

For example, say MP2 and AAC streams were both being protected by the same EEP error correction coding, and the error correction coding failed to correct a bit error in the header part of the audio frames for both MP2 and AAC. The audio would be disturbed on both MP2 and AAC and the listener would likely notice the disturbance.

It is ONLY the use of UEP that makes the DAB supporters think that MP2 is more robust than other codecs, and if MP2 were protected by EEP it would be no more robust to errors than any other audio codec.