Do they think we do not hear the difference?

[Simon B]

Possibly this is straying into areas that more properly belong on the techie board, but...

phase change rates gets more severe if you increase the number of elements of the filter

Why does this matter? Rate of phase change at a given frequency is the frequency domain dual of temporal delay at that frequency. Hence, if this is the same at all frequencies, so is the time delay at all frequencies. Under this proviso, there is no pulse distortion through the filter.

More simply, the rate of change of phase can be as severe as you like, what matters is that it is constant (2nd differential w.r.t. frequency is nil). Non-linear phase response (e.g. the case of the elementary first order RC analogue lowpass filter) leads to pulse distortion.

Given enough over-sampling (and therefore cost), i.e. effectively doing the DAC reconstruction filtering with linear-phase FIR filtering in the digital domain, it is possible to engineer a CD system with essentially perfect phase linearity, i.e. no pulse distortion.

Unless I'm mistaken, the problem Gordon is referring to arises even with a perfectly flat, linear-phase band-limited system (i.e. a "perfect" CD system). Hard band-limiting itself is the problem. Fiddling with the phase delay of lowpass filters to try to deliberately engineer pulse distortion into them is an attempt to compromise your way out of the ringing that hard band-limited systems induce on hard edges of acoustic sources (e.g. and orchestral slam) whose bandwidth is greater than that of both the recording system and the human ear. More simply - recording system=hard-band-limited, human ear=soft-band-limited, mismatch=problem.

[Gordon]

Why does this matter? Rate of phase change at a given frequency is the frequency domain dual of temporal delay at that frequency.

Yes!

Hence, if this is the same at all frequencies, so is the time delay at all frequencies.

Yes!

Under this proviso, there is no pulse distortion through the filter.

Yes! And possibly No! Depends what you mean by “distortion”. ANY change from an original state could be seen as a distortion.

If you send a square pulse [ie a classically “digital” signal that rises/falls instantaneously between 2 defined constant levels] through any band-limiting filter it will “distort” that signal by removing the frequency components that enable it to be that perfect shape. Those lost components will cause the pulse emerging to be a different shape from that which went in – so it is in a sense “distorted”. The more severe the filter shape the more the ringing.

Typically that shape for a flat, linear phase filter with very sharp roll off will tend to the well known sinx/x shaping characterised by oscillatory “ringing”. This also will be a symmetrical shaping and this is important to note because a non-linear phase shift through the filter will change that symmetry and make the output pulse lop sided because some frequency components arrive at the output “out of time” and so “distorted” in a different sense ie lose its symmetry.

Most people would consider all “distortion” as unwanted and harmful but engineers distinguish between linear distortions and non-linear ones and the latter are the harmful ones eg harmonic distortion adding frequency components that weren’t there in the input.

A linear distortion is, for example, just amplifying a signal so that its amplitude is different but an exact but larger copy! In this context shifting the phase of the frequency components such that some travel with a different DELAY through a network will introduce a linear distortion. This is generally harmless so that complex filters with a non –constant delay do not pose problems. You can almost always correct a non-linear phase response with a suitable additional network although it will complicate the network as a whole. The amount of overall delay does not matter but it is good to have it constant across the frequency range of interest. The faster the filter roll off the more that delay is.

So, provided that the input signal eg audio does not have a massive amount of energy at the higher frequencies near the filter edge the amount of removed energy is small and so the “ringing” effect is also small. However, if you hit that filter with lots of high frequency energy – percussion, loud brass, loud sharp chords on strings even, etc [heavily compressed rock music is also always loud at the top] it will ring simply because of the band limiting. The sharper that is the more the ringing. Some people can hear that ringing and if filters are too sharp they’ll hear it more.

More simply, the rate of change of phase can be as severe as you like, what matters is that it is constant (2nd differential w.r.t. frequency is nil). Non-linear phase response (e.g. the case of the elementary first order RC analogue low-pass filter) leads to pulse distortion.

Yes! See above.

Given enough over-sampling (and therefore cost), i.e. effectively doing the DAC reconstruction filtering with linear-phase FIR filtering in the digital domain, it is possible to engineer a CD system with essentially perfect phase linearity, i.e. no pulse distortion.

See above. Even in this case there will be some ringing.

Unless I'm mistaken, the problem Gordon is referring to arises even with a perfectly flat, linear-phase band-limited system (i.e. a "perfect" CD system). Hard band-limiting itself is the problem. Fiddling with the phase delay of lowpass filters to try to deliberately engineer pulse distortion into them is an attempt to compromise your way out of the ringing that hard band-limited systems induce on hard edges of acoustic sources (e.g. and orchestral slam) whose bandwidth is greater than that of both the recording system and the human ear. More simply - recording system=hard-band-limited, human ear=soft-band-limited, mismatch=problem.

Yes!! As above!! And the point about it is that it also applies to FM which also applies low pass filtering to the input audio prior to modulation. And in that case the Pre-emphasis [not used in digital systems – although there is provision in CD for it] will make matters worse by amplifying the high frequencies, doubling the amplitude every octave above 3.18 kHz. Thus the amplitude of an input at round 12kHz will be 4 times larger after pre-emphasis than it was when it went in and will tend to increase any ringing in the 15 kHz filter. That has important implications for managing FM deviation. Because these effects are linear it doesn’t matter which order the P-E and filtering is done.

[Nick_G]

Well FM only just gets to 15kHz, the transmission spec asks for better than 3 down at this frequency so there's not much point in a receiver getting much further. IOW a receiver claiming audio out to say 18 kHz is a waste of time. Pilot tone rejection has to be -40 wrt the 15 kHz spot in a receiver too so I imagine that designers will dip the audio to get that spec in with a simpler, cheaper filter. 14 kHz in DAB at say 3 down wouldn't be that much different, other things being equal. Funny how a slight lift around 3-6kHz and a touch of 2nd harmonic warms sound up too.

There's a lot to building an audio filter [the phase/transient response] than meets the ear!! It has been suggested that some of the criticism of digital sound in general is due to the sharp filters used for anti-aliassing - flat as possible to close to 22 kHz and then as fast as possible therafter. Malcolm Hawksford at Essex has done a lot of work on this and demonstrated that part of the problem is the severe transient performance that can result. Slower filters would be better but then the HiFi fan would consider the apparent loss of bandwidth a "bad thing".

DAB tricks as such in receivers are not out of the question [any more than they are in analogue] there being no definitive specification to meet. Look at any HiFi FM tuner [real ones I mean] and see how sparse the quoted specs are and how veiled the methods of measurement are. The sound of any HiFi unit isn't just down to the technology it implements as any tweaker will undoubtedly tell you. Maybe DAB uses the wrong sort of wire?

Well the manual that came with my Yamaha T-2 tuner has lots of specifications, graphs and even a schematic! It even lists what equipment was used to do the measurements which apparently conform to IHFM-2000, whatever that is. This tuner is a late 70s/early 80s model. I don't know if this was typical of tuner user manuals from that era.

I can tell you that this tuner produces the best sound from FM I have ever heard.

Regards,
Nick

[Gordon]

Well the manual that came with my Yamaha T-2 tuner has lots of specifications, graphs and even a schematic! It even lists what equipment was used to do the measurements which apparently conform to IHFM-2000, whatever that is. This tuner is a late 70s/early 80s model. I don't know if this was typical of tuner user manuals from that era.

You are fortunate than in having such details. It is not usual. The spec of that tuner is very good and it is also flexible having a choice of IF and RF settings to deal with DX and normal reception. Its S/N performance is similar to the Kenwood KT 6040 and slightly better than the Quad FM4. Beware of some of the measurements - the sensitivity relates to a 300 ohm antenna [an unfolded dipole without directors etc] which is not usual here [it needs a matching transformer for 75 ohms which will have some loss] but is common in the USA where this design was targetted - with 75 ohms there is a theoretical difference of +6dB - see the spec list [not the graph].

Strangely it seems to have negative S/N!!!??? They've plotted the graphs upside down and the DX stereo switch at about 40dBf doesn't do much. Also it refers its S/N [audio] to full deviation which is not European practice - BBC use PPM 4 as the 0dB reference which is 8 dB below the full deviation [PPM 6] and that means all the Yamaha figures need to have 8 taken off to be equivalent. It doesnt say which weighting it uses either, probably A. It claims a base audio bandwidth of 18 kHz because it has a pilot tone a cancellation system which is fine but unfortunately the transmitters are probably limited to less than that. Anyway it looks to be a very good performer, but then it's a Yamaha whose products by and large are well designed and made.

Notice it needs about 1-2 millivolts from the antennna to give of its best [10 for stereo] so a good antenna is a must. As we keep emphasising, good FM needs a good input signal which means a good antenna. Its mono hits threshold at about 5dBf which is about where the Kenwood and Quad are too and stereo seems to be heading for 15 dBf.

The use of old [probably obsolete] reference standards is interesting. The US IHFM [Institute of High Fidelity Manufacturers] was similar to the old DIN 45500 standard now somewhat neglected because it has fallen behind the times. No audiophile would consider the IHFM or DIN standard adequate these days but it did a good job back in the early days.

[NHTL]

I so agree with JLW's comments. I listen via an Arcam Neo and a Quad FM4 either via Monitor speakers or Beyerdynamic DT 990 headphones. I find that I can hear the difference between 192 Kbps discrete stereo and 160 Kbps Joint Stereo, despite being 70. As has been mentioned above, In JS I am aware of the less stable stereo soundstage and the less vibrant sound. I tend to listen more via DAB as I prefer the greater dynamic range that it offers. I have an excellent DAB and FM signal (I can see the transmitter from my living room window). If anyone cannot hear the difference between the sound of FM and DAB I suggest that they listen not only to the sound of the strings, but especially to the sound of clapping. Switch as quickly as you can between FM and DAB when there is clapping. Let us hope that DAB + is eventually adopted in the UK at a reasonable bit rate. I recently heard classical music on DAB + in Germany and was impressed by the sound quality.

[Nick_G]

You are fortunate than in having such details. It is not usual. The spec of that tuner is very good and it is also flexible having a choice of IF and RF settings to deal with DX and normal reception. Its S/N performance is similar to the Kenwood KT 6040 and slightly better than the Quad FM4. Beware of some of the measurements - the sensitivity relates to a 300 ohm antenna [an unfolded dipole without directors etc] which is not usual here [it needs a matching transformer for 75 ohms which will have some loss] but is common in the USA where this design was targetted - with 75 ohms there is a theoretical difference of +6dB - see the spec list [not the graph].

Strangely it seems to have negative S/N!!!??? They've plotted the graphs upside down and the DX stereo switch at about 40dBf doesn't do much. Also it refers its S/N [audio] to full deviation which is not European practice - BBC use PPM 4 as the 0dB reference which is 8 dB below the full deviation [PPM 6] and that means all the Yamaha figures need to have 8 taken off to be equivalent. It doesnt say which weighting it uses either, probably A. It claims a base audio bandwidth of 18 kHz because it has a pilot tone a cancellation system which is fine but unfortunately the transmitters are probably limited to less than that. Anyway it looks to be a very good performer, but then it's a Yamaha whose products by and large are well designed and made.

Notice it needs about 1-2 millivolts from the antennna to give of its best [10 for stereo] so a good antenna is a must. As we keep emphasising, good FM needs a good input signal which means a good antenna. Its mono hits threshold at about 5dBf which is about where the Kenwood and Quad are too and stereo seems to be heading for 15 dBf.

The use of old [probably obsolete] reference standards is interesting. The US IHFM [Institute of High Fidelity Manufacturers] was similar to the old DIN 45500 standard now somewhat neglected because it has fallen behind the times. No audiophile would consider the IHFM or DIN standard adequate these days but it did a good job back in the early days.

Thanks Gordon. I can't say I've noticed any loss using the 75 ohm cable hooked up to the Yagi. It's very sensitive, comparable to my other tuners, and the selectivity in narrow is very good as well. In fact the specs suggest it should be Yamaha's best DXing tuner. I'm also amazed that such a densely packed tuner of that age runs so cool - it barely gets warm even after several hours.

Interesting that Yamaha provided so much information in their user manual. I just thought that perhaps 70s tuners had more specs & graphs in their manuals as it was the time when tuners were an important source component and there was much competition between rival Japanese, US and European manufacturers.

Regards,
Nick

[Bryn]

If you want to get some idea of what DAB+ is likely to sound like if the Beeb eventually adopts it, try listening to Radio 1, 2 or 4 via the iPlayer with High Bandwidth selected. What you get is around 128kbps HE-AAC, which is what BBC DAB+ would be most likely to use. If they are kind enough to leave Radio 3 at 192kbps, the quality will be somewhere between the current high bandwidth offering for Radios 1,2 & 4 and the current HD Sound version of Radio 3 (320kbps AAC-LC).

[An_Inspector_Calls]

More simply, the rate of change of phase can be as severe as you like, what matters is that it is constant (2nd differential w.r.t. frequency is nil). Non-linear phase response (e.g. the case of the elementary first order RC analogue lowpass filter) leads to pulse distortion.

I didn't make myself clear.

Sticking to low-pass filters, in any single order filter the frequency response will start to change at the design frequency, be 3 dB down at that point, and be asymptotic to a 6 dB/octave response change. At the rollover point the phase change will be 45 degrees and eventually attain 90 degrees. Thus, about the rollover frequency there's varying phase changes being applied to the audio signal: phase distortion thus occurs at least as far down as one octave below the rollover frequency. In the case of LP equalisation I stated that there's a constant phase change applied across almost the entire audio frequency spectrum by the equalisation filter. In the frequency domain there's a 6 db/octave roll off; in the phase domain there's a constant 90 degree shift. On that basis there won't be any phase distortion; square waves will be well handled by LP.

In the case of CD there's a filter created by the 44 kHz sampling rate which will apply an abrupt frequency cut at 22 kHz. This will have a roll off rate far steeper than a single order filter and of course make for a phase change at frequencies above 22 kHz and is bound to create phase changes below as well. But because the CD filter is so steep, many octaves below the 22 kHz roll-off there'll be significant and varying phase changes. Can this be corrected by applying many high order filters of opposite effect? As for correction by over sampling, is there any proof this will correct the phase distortion? Hence, across the frequency spectrum CD is unlikely to be phase coherent. Transients will not be handled very well by CD.

[Gordon]

... If anyone cannot hear the difference between the sound of FM and DAB I suggest that they listen not only to the sound of the strings, but especially to the sound of clapping. Switch as quickly as you can between FM and DAB when there is clapping. Let us hope that DAB + is eventually adopted in the UK at a reasonable bit rate. I recently heard classical music on DAB + in Germany and was impressed by the sound quality.

Chances of DAB+ terrestrial broadcasting in the UK are not at all good. As Bryn
suggests you can get it via the web site.

Clapping is very similar to white noise and so any compression system is going to struggle, there's no structure to get hold of. There is no useful psycho-audio model to help steer the quantisation. With MP1L2 the sub bands will all be equally filled with energy demanding bits. similarly DAB+ will still suffer with noise like signals.

That gives us a clue as to why certain sounds cause trouble - could it be that anything that produces a wideband noise like spectrum will spread available bits too thinly? I find that speech sibilance [s's and t's] does sound odd but then that is emphasised by announcers being close to their mics and it happens on FM too.

Compressing noise [a task for the insane] is very interesting because theory suggests that the process should take noise in and put whiter noise out by adding its own quantisation noise to the mix [central limit theorem]. So "pink" noise in - whiter noise out?? Perhaps the difference you are hearing in clapping [pink noise] in particular is inevitable and that the compression is "purifying" the noise!!

[Gordon]

Thanks Gordon. I can't say I've noticed any loss using the 75 ohm cable hooked up to the Yagi. It's very sensitive, comparable to my other tuners, and the selectivity in narrow is very good as well. In fact the specs suggest it should be Yamaha's best DXing tuner. I'm also amazed that such a densely packed tuner of that age runs so cool - it barely gets warm even after several hours.

Interesting that Yamaha provided so much information in their user manual. I just thought that perhaps 70s tuners had more specs & graphs in their manuals as it was the time when tuners were an important source component and there was much competition between rival Japanese, US and European manufacturers.

Regards,
Nick

You have a good signal where you are and have a decent antenna so your antenna mismatch will not lose you much in perceived performance.

The S/N measurement of the noise is likely to be in a bandwidth less than the claimed audio ie 15 instead of 18 they don't say. One other thing about that spec = the distortion figures are very low, suspiciously low, Note the log frequency scale - the low distortion is all for low audio frequencies where bandwidth limits don't affect things - at 1 KHz there are at least 180 sidebands available and at low deviation bandwidth isn't under pressure anyway. At 10 kHz that is not so and distortion rises rapidly especially at high deviation. These distortion figures will also include harmonic and possibly noise as well so must be done at high RF input [65dBf]. Depends how it's done - again that IHFM spec isn't to hand. The HP4333A may be a true harmonic analyser. EDIT: it isn't, it's conventional and so does measure THD + noise.

I suspect they are measuring [they don't say and I can't find that IHFM spec anywhere on the web] with very low deviation. The rise at the top of the band and with low RF input is to be expected.

[David-G]

Joint stereo [JS] is where the true separate R and L signals are not coded independently although there is obviously a lot of similarity between them a lot of the time. That similarlty is exploited in JS so that the stereo effect only gets rendered vai the loudness of the R and L signals. Any time of arrival differences at each ear are largely lost and so there is a degradation in the perceived stereo image or soundstage. The bits that are saved in this process are used to give to the coding of tonal/spectral information and that "justifies" a drop in bit rate!

It's available for use in most audio compression systems. EDIT: In principle, the more granular the compression algorithm is and the more complex and analytical it is the better it will do the job in JS. The MP2 algorithm is not really sophisticated enough to do an excellent job in JS mode. The effects that Bryn highlights above will be less apparent if spectrum analysis is finer.

Thanks Gordon, that's very helpful.

[Nick_G]

You have a good signal where you are and have a decent antenna so your antenna mismatch will not lose you much in perceived performance.

The S/N measurement of the noise is likely to be in a bandwidth less than the claimed audio ie 15 instead of 18 they don't say. One other thing about that spec = the distortion figures are very low, suspiciously low, Note the log frequency scale - the low distortion is all for low audio frequencies where bandwidth limits don't affect things - at 1 KHz there are at least 180 sidebands available and at low deviation bandwidth isn't under pressure anyway. At 10 kHz that is not so and distortion rises rapidly especially at high deviation. These distortion figures will also include harmonic and possibly noise as well so must be done at high RF input [65dBf]. Depends how it's done - again that IHFM spec isn't to hand. The HP4333A may be a true harmonic analyser. EDIT: it isn't, it's conventional and so does measure THD + noise.

I suspect they are measuring [they don't say and I can't find that IHFM spec anywhere on the web] with very low deviation. The rise at the top of the band and with low RF input is to be expected.

Thanks Gordon. I am also into long-distance reception (DXing) and the T-2 is very good at eking out very weak signals, again with no obvious loss of signal compared to my other tuners. One tuner which is a tiny bit better (very slightly lower levels needed to reach full quieting in mono) is the Denon TU-800L but this is because it has a 3rd 'super narrow' bandwidth which is slightly narrower than the narrow mode in the T-2. Narrowing the IF raises the apparent sensitivity at the expense of greater distortion.

Maybe the distortion figures are correct. I do know that with a strong signal on a good broadcast is sounds more musical and true-to-life than any other tuner I've heard.

Regards,
Nick

[Gordon]

Maybe the distortion figures are correct.

I'm sure they are! I just wanted an idea how they were measured because they are so low. I would have preferred a set done at say 30 kHz, a deviation more representative of practical use. They do quote 40kHz in the "usable sensitivity" spec.

The "distortion" [aka THD + Noise from the HP433A] at 0.01 %[see graph in #25] is 0.0001 of the input, if referred to 1 Volt [to be comparable to the S/N], is 80dB down in amplitude [assuming that the 0.01% is amplitude not power] and presumed to be dominated by the distortion not the noise. How does that square with the S/N at 85dB down?? The distortion [measured at the same time with the same device] needs to be smaller than the noise. IOW the "noise floor" is in fact set by irreducible distortion not noise.

Deviation = amplitude = reference signal level used for measurement. We have no declared ref level but if it is 1 volt then the deviation must be 75 kHz and so the S/N will be dominated by distortion. The measurement deviation must be about 1 tenth of 75 to avoid this, but what is it actually?

Compare the graph of #21 with this:



See that the RF level dropping appears to increase the noise as expected: at RF input 15dBf the stereo " distortion" is heading for >5%, possibly 10, but the S/N graph gives a figure of about -30dB. If that 5% is the amplitude wrt 1 volt this is 50 millivolts, 0.050 volts or -26dB so the "noise" contains a considerable amount of distortion too. 10% would give the same reading as for S/N. At 35dBf the distortion is at 0.5% now 10 times or 20dB better so at -46dB and the S/N is now -51dB so now we can say that the "noise" has distortion in it. When we get to-65dBf input the distortion is now 0.033% or -0.00033 wrt 1 volt or -70dB whereas the S/N is -78dB - how come??? The S/N should bottom out at the distortion level.

[NHTL]

Chances of DAB+ terrestrial broadcasting in the UK are not at all good. As Bryn
suggests you can get it via the web site.

Clapping is very similar to white noise and so any compression system is going to struggle, there's no structure to get hold of. There is no useful psycho-audio model to help steer the quantisation. With MP1L2 the sub bands will all be equally filled with energy demanding bits. similarly DAB+ will still suffer with noise like signals.

That gives us a clue as to why certain sounds cause trouble - could it be that anything that produces a wideband noise like spectrum will spread available bits too thinly? I find that speech sibilance [s's and t's] does sound odd but then that is emphasised by announcers being close to their mics and it happens on FM too.

Compressing noise [a task for the insane] is very interesting because theory suggests that the process should take noise in and put whiter noise out by adding its own quantisation noise to the mix [central limit theorem]. So "pink" noise in - whiter noise out?? Perhaps the difference you are hearing in clapping [pink noise] in particular is inevitable and that the compression is "purifying" the noise!!

Thank you Gordon for this explanation about clapping, it makes a lot of sense. Does this mean that the Nicam distribution chain to the FM transmitters can cope because it is a much less lossy system?

[Resurrection Man]

Thank you Gordon for this explanation about clapping, it makes a lot of sense. Does this mean that the Nicam distribution chain to the FM transmitters can cope because it is a much less lossy system?

Unless I am mistaken, NICAM was used on TV sound....nothing at all to do with FM.