DAB Report

[Bryn]

O.k., I recorded/saved last night's Mahler 6 from FM (Wrotham - good strong signal to Sony D777ES and on to Alesis Masterlink at 44.1kHz sample rate and 24 bit quantization), DAB (the mp2 stream saved as a file, then resolved to WAV with WinAmp) and the iPlayer HD Sound FLV file from which the AAC-LC file was extracted with FLVextract.exe and converted to WAV with Free m4a to mp3 converter, which has a convert to from aac to wav option). Once the FM WAVs are transferred from the Masterlink to this laptop, I will use Sony Sound Forge Pro 11 to normalize all three to the same average RMS level, then produce simple x/y line spectrographs of each and post them here, along with the peak and average RMS stats for each, the latter to give an indication of the relative dynamic range of the different modes of transmission.

[Gordon]

O.k., I recorded/saved last night's Mahler 6 from FM (Wrotham - good strong signal to Sony D777ES and on to Alesis Masterlink at 44.1kHz sample rate and 24 bit quantization), DAB (the mp2 stream saved as a file, then resolved to WAV with WinAmp) and the iPlayer HD Sound FLV file from which the AAC-LC file was extracted with FLVextract.exe and converted to WAV with Free m4a to mp3 converter, which has a convert to from aac to wav option). Once the FM WAVs are transferred from the Masterlink to this laptop, I will use Sony Sound Forge Pro 11 to normalize all three to the same average RMS level, then produce simple x/y line spectrographs of each and post them here, along with the peak and average RMS stats for each, the latter to give an indication of the relative dynamic range of the different modes of transmission.

Thanks Bryn!! You deserve a or two after all that. Look forward to seeing the plots.

[Bryn]

Here we go. Slight change of plan. Normalizing to the same average RMS value proved to be impractical, so they are each normalized to -1dB peak level.

FM (Average RMS -18.3dB)



Somewhat untidy filtering of the 19kHz stereo multiplex pilot tone, but the peak is much less prominent that that on the Testament issue of the Boult Havergal Brian Gothic.


DAB (Average RMS -21.8dB)



Clearly they have indeed installed updated mp2 encoding since my last look a few years ago.


HDS (Average RMS - 24.8dB)



Mmmm. Nice!

[for those not technically minded, the RMS value are only an indication of the relative differences in dynamic range. Don't confuse them with absolute values for the ranges. I don't have the wherewithal to measure those directly.]

[Gordon]

Thanks Bryn, sterling work there, definitely worth a few . And I think you are right to normalise to a peak related value. I'll need some time to absorb these plots but it is very obvious that FM's RMS power is higher than the others suggesting that the quit bits are not so quiet!! That implies that the DR is a lot less than the other 2, probably the Optimod in action. But why would you limit the DR of FM and not DAB? The control is meant to help in car and portable listening and varies in the day so that this concert should have had better DR than perhaps during drive time. There is a DR control in DAB of course but AFAIA no-ne uses it.

The FM pilot tone region of your Sony 777 [a very good receiver BTW] is a bit weird. Not sure how to read that area, it's clear that the bandwidth is being limited at around 15 kHz, by the transmitter rather than the receiver which probably offers more if it can get it. The equivalent Kenwood certainly does.

These are of course spectrum density plots averaged over the 80 seconds of the clip. To get a total signal power measure one would need to integrate the plot over the band [that's the RMS]. To get a true DR meaure we'd have to see a bit of max signal power and some of lowest and integrate those across the band too. This would be visible in the time plots.

Notable is the amount of power below 100Hz and more so 10! Short term variations are not shown, eg a few seconds of loud and active and perhaps also a few seconds of something quieter and less so, but that wasn't the intention. The clear trend is for the HF region beyond the spectral bulk area [say 100 - 500 Hz, where all the major fundamentals are] to decline in average power density by about 40dB by 5kHz and another 40 or so by the top of the band at about 21kHz [looking at the HDR feed]. Enough for now.

[Gordon]

Too many points to cover in full so I will try to be selective.

Ditto. I think we have to agree to differ on a number of points.

I do not deny what your ears tell you, it doesn’t align with what mine tell me. But then we have not sat in the same room listening to the same things for a while have we? The discrepancy is what intrigues me. Your dislike of DAB sound is vehemently expressed implying a very unpleasant experience. Why bother with DAB at all? Why own any DAB receivers?

If two systems sound very similar but I can hear differences and then form a view as to my preference would I then praise that preference and label the other as execrable? I wouldn’t. In the case of DAB I don’t believe that it is so bad as to attract the level of your criticism especially in the listening conditions you describe. Perhaps your hearing is more acute than mine or perhaps I’m just not as fastidious about sound? Or perhaps we aren’t listening for the same things.

You can relax now anyway because there will be no forced switchover for many years to come.

[Bryn]

To respond to Nevralti's enquiry Re. the use of discrete stereo. As far as I am aware, the only UK DAB channel to use it is BBC Radio 3 when at its standard 192kbps data rate.

When I get home from work this afternoon, I plan to seek out the quietest couple of seconds of ambient at identical points in the three Mahler 6 files and report the average RMS level of each. That should provide a more easily graspable representation of the relative dynamic ranges of FM, DAB and HD Sound.

[It would appear that Sound Forge has a problem with measuring the average RMS of short, very quiet, clips. So I have had to abandon the above project.]

[Dave2002]

I do not deny what your ears tell you, it doesn’t align with what mine tell me. But then we have not sat in the same room listening to the same things for a while have we? The discrepancy is what intrigues me. Your dislike of DAB sound is vehemently expressed implying a very unpleasant experience. Why bother with DAB at all? Why own any DAB receivers?

Some sensible points here, as usual.

We actually fall into the group who listen to DAB frequently, as we have a set in our kitchen, and we use it for listening to the Archers, News, Desert Island Discs etc. For what it does it's generally fine, though sometimes we get the bubbling mud and severe distortion. I think that has probqbly been less since the TV switch over.

We also listen to FM, to TV sound and to Internet versions of R3 and R4.

Your mention of the Sony 777ES tuner reminds me that it might be worth my while reconnecting mine, and maybe also my JVC tuner FM tuner - both very good.

Re listening - a lot depends on circumstances. Last night I listened to some Mozart courtesy of another board member and noticed some low frequency background noise. However, when I listened on a set of headphones - generally OK but presumably without the bass response even of my small speakers, the background noise disappeared - and presumably a lot of the low pitched instruments too.

Different people have different hearing characteristics. I have heard some unpleasant sounds from DAB in the past, and in fairness since I don't normally listen to it via the same equipment that we use for CDs and for some listening via Internet I can't say how poor it is currently. I never rated FM that highly, but a former board member sent me a recording he had made from his FM tuner, and it made me rethink all over again. He had managed to get really low hiss levels, something which I've not done since moving house over 10 years ago - though that's simply indolence on my part!

It is unfortunate that DAB is limited by the bit rate and the codec used. At 256kbps or higher the codec, though not optimal, should still give good results, and that was how at one time DAB was intended to be used.

[Nevalti]

.......DAB (Average RMS -21.8dB)

Interesting plots; what do they sound like though?

I note the obvious DAB drop-off above 10,000Hz compared with the others. Do you think this is a real and deliberate level reduction to mask a problem or is it really a reflection of the rejection of low level information considered by the codec to be 'noise'.

Whichever it is, it seems to support my perception of one of the problems with DAB. It will be interesting to see your 'close-up' comparison. I wonder if that will also show information missing on DAB.

Is it feasible to also look at and compare a simple wave form?

[Bryn]

Interesting plots; what do they sound like though?

...

Here is a link to a sequence of the the close of the performance in each of the three transmission modes. It should be obvious which is which:

http://we.tl/jwTX3gmtv8

The DAB has been re-sampled (with anti-aliasing filter) to 44.1kHz and the FM noise shaped and dithered to 16 bit (both procedures using iZotope tools.

As to waveforms, those, for the whole symphony, would not be a practical proposition to post here. I could post the final quiet 'bomp' though, if is would be thought useful.

Here's the 'bomp':

[Gordon]

Interesting plots; what do they sound like though?

Sorry this post is going to be bit long but if we’re going to get anywhere…

Hmm, yes, the more I look at these the more there is to think about. Good question. If the droop is real then at first sight it should be audible at least on decent equipment. The next question is: Is this droop consistent with your experience of DAB listening? Is the loss of HF “brightness”, which a tweak of the treble tone control could fixed, if we had one, what you hear and what causes your antipathy towards it? If so we may be getting somewhere. If not, there is something else we are not seeing.

Given how low the density is up there at the top, -60dB or more cf 500 Hz, even though it looks to droop, it is a very small proportion of the total signal energy which is much, much greater down below. So how much influence this might have subjectively is hard to say without listening. Casual listening on average equipment may not reveal it. It is possibly the capacity to detect this sort of small HF energy difference that separates people. Hearing HF tones individually is one thing hearing them in a complex spectrum is another.

Anyway, to the plots. First, what exactly are we looking at?

If I understand Bryn correctly [see #91, 93] the 3 plots are taken from the same 80 seconds of the concert. From them we can't say what the dynamics were during those 80 seconds because we'd need to see the time version as well. Nevertheless they are the same for all 3 channels. It would be useful if Bryn could be prevailed upon to post one waveform of the 80 seconds – we donlt need them all or even both channels.

There are DR differences apparent and it's good that the anchor for normalisation is the peak amplitude in each case. The absolute values of the RMS don't mean much because of these DR differences and because we don't know what proportions of time the music captured spends at high or low levels. The more time it spends at low level over these 80 seconds the lower the RMS value and vice versa. Similarly the less it spends there the lesser the DR will be for a given RMS. IOW the differences we see in the RMS values will be exaggerated the less quiet music there is. In the FM case we don't know what, if anything, the Optimod is doing over these 80 seconds. If there are longish periods of loud then soft then perhaps it will have an effect.

Secondly, these are spectrum density plots and we don't know the resolution, unless the software declares it somewhere, but no matter it is the same for all 3 plots and so comparing the shapes of the plots is valid. The resolution bandwidth only affects the sit of the plots in the vertical direction. These are NOT frequency response plots, they show the power residing in each of the FFT bins and reflect the spectral distribution of power in this particular 80 seconds of music, any short term deviations will be swamped.

The coarse vertical scale of the plots does not allow any precise estimates of the respective spectral shapes.

I am still concerned about what the plots are doing below say 100 Hz. Where is that spectral power coming from? It being live, I suppose there could be air con in the background or general rumbling but ... Also that strange stuff around the pilot tone is very odd.

I note the obvious DAB drop-off above 10,000Hz compared with the others. Do you think this is a real and deliberate level reduction to mask a problem or is it really a reflection of the rejection of low level information considered by the codec to be 'noise'.

See this plot



which I have cut & pasted to make the 3 systems easier to see. I’ve shifted the plots to get the 10KHz response about the same and added some cursor lines to help see what’s what. That droop is real enough and it doesn’t come from the source material, it clearly is a feature of the DAB channel. Now what is the cause?

As you suggest, it could either be a deliberate attempt to assist the codec [see the second para of my post #88 above] by reducing the HF content OR it could be the algorithm itself. There is compression in the HDS stream of course but it doesn’t exhibit the same behaviour, largely because it is a better codec and also there is a lot more bit rate.

Now we need to be careful. Insofar that all 3 plots have been rendered identically, the fact that there is clearly less power in the top bins in the DAB case means that something wasn’t there to begin with or it has been filtered out in the channel, but what did that? One assumes that the DAB receiver is blameless and doesn’t have that droop.

So what does MP2 do that could effect this degree of loss? The sub-band filter set is coarse – probably much coarser than this spectrum resolution is – and so will apply its masking process to filter out “inaudible” components in the sub-bands. If the algorithm is automatically selective under control of the psycho-acoustic modelling this reduces the amount of residual power differently and dynamically in each sub-band. This would probably appear as a variable level drop all across the board, not just at HF. IOW this droop could vary depending on the material being coded.

If we can observe a greater loss in, say, the R2 feeds that use much lower bit rates and JS then perhaps we might get a clue? I find it difficult to find a definitive reason to explain that droop in FR, real as it appears, but I’m not ready to blame the algorithm yet but it has to be a suspect.

Whichever it is, it seems to support my perception of one of the problems with DAB. It will be interesting to see your 'close-up' comparison. I wonder if that will also show information missing on DAB.

As noted above, the FR droop should be heard as a softening of the sound, a loss of brightness. This loss should be more apparent in a decent HiFi system whose HF is extended and clean. However, on less capable systems who knows?

The “close ups” I was hoping for would reveal the DR of the clip we‘ve analysed. IOW a section of waveform with a peak and a trough in it. The files that Bryn posted are however a good substitute because we can hear them and ourselves get a waveform. I’m still listening to them [Thanks Bryn!, that’s yet another beer we owe you] - BTW I assume that hese are not the clips that were analysed above? But here are some plots taken from that file: the two images should be self explanatory:

Is it feasible to also look at and compare a simple wave form?

Se above although I’m not sure what you mean by simple. Without getting the BBC to send out some frequency sweeps I don’t see how we can get at the information we need. Anyway, if you give the MP2 process a single frequency, or even a small set of them, it will use its bits on only them and render them very well indeed. So, if this droop is caused by the algorithm and not by a deliberate filtering, then it should disappear with tone tests. The result of simple tone testing is exemplary figures for S/N, harmonic distortion, etc. That is why testing compression systems using analogue methods – ie sine waves in audio - doesn’t help. Analogue tests are not at all useful for compressed systems. I’m not sure what we can make of the waveforms of the “bomp”.

[Gordon]

Just to add more intrigue, here are some spectrum plots taken using Audacity. They are the spectra of the three segments shown in the last 2 waveform images of the post above and taken from the last 20 seconds or so of each excerpt and listed in order from top to bottom. They have not been normalised and the amplitudes are those inherent in the file that Bryn sent earlier for us to listen to. You can see from those waveorms above that the peaks in each case or very similar. Click on each image and it takes you to imgur where it is stored and you'll be able to expand it yourself.



One obvious thing is that the resolution is much finer showing more details of the spectral distribution. Using some clear harmonic lines in each spectrum, and the cursor of Audacity to give the frequency, here are some amplitude measures for each of the 3 plots [tabbing doesn't work!]:

[Nevalti]

..........I find it difficult to find a definitive reason to explain that droop in FR, real as it appears, but I’m not ready to blame the algorithm yet but it has to be a suspect...........

Agreed, there could be several reasons, most of which I would probably not understand.

Take away or reduce HF and the 'air' or 'life', the 'acoustics' information goes out of music making it sound dull. Take away minute details, such as decaying reverberation, and the basic notes become very clear but also sound artificial because(?) the harmonics of those notes have been chopped out. I believe that is why DAB can sound commendably clear but ultimately unrealistic. We suffered similar issues in the early days of digital recording where far too much low level information was discarded as 'noise'. Engineers were so pleased to be rid of tape noise that they sought a totally silent backgrounds by filtering out all low level info. Impressive - until you realised that it sounded artificial - because it was artificial.

We can hear the same effect now when comparing a DAB or Freeview transmission with an FM or internet (320kbps aac) R3 transmission. Listen to a live transmission from somewhere large like a cathedral and you can actually count the echoes from a distinct note. It can be very obvious that the echo continues longer with FM and 320 kbps aac because(?) the low level information is not so obviously chopped off - if at all. That shortcoming, and to me it most definitely is a shortcoming of DAB, seems to be simply a function of the limited capabilities of the codec. It is old, basic, severely limited in abilities and drastically outdated.

The solution is blindingly obvious yet we still have people trying to justify persevering with this relic.

[retroman]

The solution is blindingly obvious yet we still have people trying to justify persevering with this relic.

Gentlemen, need he say more?

[Gordon]

....most of which I would probably not understand.

Not necessarily. I don't understand where that droop came from and, until we understand the evidence we can't form conclusions. As I have said above, I'm still thinking about it which is why I did the spectrum again using Bryn's file. It implies a different story. Why?

....limited capabilities of the codec. It is old, basic, severely limited in abilities and drastically outdated.

Yes, compared to other methods of coding, it is, if you judge it by audiophile standards of expectation and listening intensity, we all know that!! You are clearly a person who has those high standards, but I fear you are in a minority who listen like that. I'll repeat what I have said before in this thread: DAB as it stands today is not, was not and never will be an audiophile medium. What imperative is there that it should be?

The solution is blindingly obvious yet we still have people trying to justify persevering with this relic.

If you mean moving to DAB+ with decent bit rates then would I agree. It has been suggested to government and others for some years that DAB+ should be adopted but there is a solid brick wall erected by an industry caucus who do not want change, and don't feel it is justified to please a tiny minority of audiophiles. Regrettable though it may be, it is a simple as that. I know, I've felt that brick wall.

Some months ago this debate ran here, yet again, as it has before in places like Digital Spy, and there was lots of carping and griping but no-one was prepared to take action ie form a pressure group to bombard newspapers, HiFi magazines, government, MPS, radio manufacturers, broadcasters, etc etc with complaints and suggestions for improvement in the form of a plan for a transition. It would involve leadership, organisation, probably raising some cash to fund it and a lot of time and effort - surely it's worth it? Turn all that passion into action? People have tried and have not succeeded but that doesn't mean it is impossible. Who is prepared to put actions where their mouths are?

[Bryn]

The spectra I posted were derived from the full 80 minutes of the performance, not just 80 seconds. The audio samples were kept short to save on data.