[Premium-Rx] Re: Phase Linear Filters

[Michael O'Beirne]

Hi Gents,

This is my 1 Euro's worth to add to George's piece.  He is absolutely right in that the phase response of a filter depends on the type of filter, but there are other factors too, for example the Q Factor of the filter.  In the case of a double-tuned circuit (as used in the traditional IF transformer) the Q factor defines the shape of the selectivity curve, though it will have a Gaussian response and give hardly any phase distortion.  As the Q is raised by using crystal elements or mechanical filters, the slope of the filter will improve but the phase distortion will increase.

For some reason, the transient response of early mechanical filters was a lot worse than that of crystal filters.  An academic reference is the paper: "Survey of Mechanical Filters and their Applications" by J C Hathaway and D F Babcock - Proc IRE, January 1957.

For PR's non-technical readers, the phase response is a measure of how linearly in time all signals pass through a filter.  The linearity is usually good in the centre of the filter but deteriorates badly towards the filter edges.  This means that the signals towards the LF and HF ends are substantially delayed compared with signals in the centre of the passband.  For voice signals this does not matter too much but for data signals it can be a disaster because the data recovery depends entirely on the relative timing of the pulses.  Furthermore, pulses with fast rise and decay times are distorted particularly badly.  (Claude Shannon's equations and all that theory).  That is partly why the rounder voice signals are less affected and why some data systems deliberately round over the edges of the pulses, as in "Clover".

If you want to send fast FSK at HF, say at 1200Bd, you need to use special crystal filters with phase compensation so that the group delay through the filter is equalised - which means that nearly all the signals emerging after the filter have retained their relative phases.  An ordinary filter will cause bad data distortion.  

Such phase compensated filters tend to be very expensive and often have to be hand built.  I have a written quote for the phase compensated LSB filter for my Marconi H2540 receiver at £1250 plus Value Added Tax (I joke not).  I acquired a matched USB/LSB pair for £10 at a flea market because the vendor had no idea what they were.  They have big cans, about twice the physical volume of a normal professional crystal filter.  The 4dB points are quoted (with respect to the final IF) at +300Hz and +3,300Hz, and the 60dB points lie at -300Hz and +4,000Hz.  Having fitted them in lieu of the original filters, there is a noticeable improvement in the quality of the recovered audio, particularly listening to an AM station in SSB mode.  In contrast, doing this with a ham receiver and a 2.1kHz filter gives a pretty dreadful sound quality.  The disadvantage is that a little more QRM and noise creep in at the HF end, but the 60db bandwidth is better.  Curiously, the measured s+n/n ratio improved a little with the new filters despite the wider 5dB points.  This may be because the overall area contained within the selectivity curve had reduced.

A test I use on all receivers (which requires no test gear) is to listen to the BBC's Radio 4 Long Wave on 198kHz, which is 90dBuV here.  The modulation used on the BBC transmitter is a special switching method called "Optimod" which a design engineer told me is more efficient than traditional methods and gives 100% modulation.  If you listen carefully via decent hifi headphones on a quality receiver you should hear the clipping of the syllables.  You won't hear this with a poor receiver because of poor phase response.  Another simple test is to listen to a scratchy AM station on HF in AM and then SSB.  I use the BBC on 12,095kHz.  My QTH is well into the skip zone and the signal is weak here, sometimes into the noise.

Good phase response is also relevant for very tight filters for CW.  A super selective filter will separate the stations but be very tiring to listen to after a while. Thus the 300Hz filter in the RA1792 has a relatively poor 6 to 60dB shape factor compared with the published responses of some ICOM filters but it is very easy to listen to over a lengthy period.

DSP has the great advantage that it is inherently a low Q arrrangement achieving both excellent shape factors and good phase linearity.  But there are problems (so far).  The first is that inevitably the DSP is achieved at the back end of the IF chain - round about 30kHz - whereas the best place for a sharp filter is immediately after the first mixer - as was done with some very "un-PR" receivers such as the Drake R4C with the Sherwood mods.   The second problem is the limited linear range of current A-to-D converters.  This requires AGC-controlled analogue amplifiers ahead of the digital stuff to limit the signal level to the A-to-D converter.   The third problem is the sizeable delay of the DSP in responding to a strong signal.  This can lead to a short-lived overload on a strong signal while the DSP gets to grip with the strong signal.  Finally, all manner of instrusive aliassing noises can be added by the processing.  The day will eventually come when DSP can be achieved at 40MHz and then we should have superb rigs.

End of the 1 Euro lecture.

73s to all,
Michael
G8MOB
----- Original Message ----- 
  From: George Georgevits 
  To: Premium-Radio 
  Sent: Saturday, January 15, 2005 9:18 PM
  Subject: RE: [Premium-Rx] Siemens CHR 531 and Phase Linear Filters


  Greetings to all,

  Just to add my $0.02 worth, I always thought the phase linearity of a filter was not dependent on how it was physically realised (LC, active, mechanical, quartz crystal etc.) but rather on its design response type, ie. whether it is a Butterworth, Bessel, Tschebyshev, Elliptical, to put a name to the most common types of filter designs? Each filter design type can then be applied to a low pass, high pass, band pass and band reject response, as the application requires. 

  Each of these design responses has its own magnitude and phase response characteristics. Bessel filters have the most linear phase response, but the poorest magnitude cutoff characteristics. Consequently, they require many more cascaded stages (or orders, as they are called) to achieve the same cutoff slope as an equivalent elliptic function filter (which, by the way, has the worst phase response of the types mentioned above).

  Regards,
  George Georgevits



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.qth.net/pipermail/premium-rx/attachments/20050116/7cd071f3/attachment.htm