GM3WOJ - GM2V - ZL1CT website March 2011
Section 1 - HF RX performance information from Geoffrey Mackenzie-Kennedy GM4ESD
Section 2 - INRAD and Elecraft roofing filter plots from Gavin Taylor GM0GAV
Section 3 - Comments from Dave Johnstone GM4EVS
Section 1 - the following notes are compiled from a series of e-mails from Geoff GM4ESD
During the past two decades (roughly) there has been a "horse race"
between Direct Sampling Software Defined Receivers and Superhet Receivers, the
being which platform offers the better performance in terms of 3rd Order Dynamic Range and Noise Figure. So far the superhet platform is still ahead,
but some direct sampling receivers are closing the gap. At this time a modern design of superhet can yield a larger dynamic range and lower Noise Figure
than that of the direct sampling SDR- for example an IMDDR3 of >120db at 2kHz spacing using a 500 Hz roofer and > 110db with both tones inside the
passband, together with a HF Noise Figure of <10db for the superhet without a preamplifier.
The quality of crystals used in the superhet's roofer, and their matching, is of paramount importance.
From a paper analysis of the K3's receiver, the IMDDR3 would appear to "droop" from < 95 db ( @ 2 kHz spacing) to approximately 70db - 75 db when
both tones are within the passband, i.e. a "droop" of roughly 20 -25db from the delta 2 kHz IMDDR3. This compares with a "droop" of approximately 10db
from >120db to > 110db as seen in a modern superhet, the result from the modern design is not phase noise limited
Some people would argue that a large dynamic range (IMDDR3) within the passband is not useful, because key clicks and/or phase noise generated by
transmitters operating on close frequencies would become the dominant problem. I cannot agree with this argument after using receivers whose
in-passband IMDDR3 is large for the past 16 years, which includes using them for working SSB DX on 40m from this Scottish QTH amongst the BC stations
whose carrier levels could reach +5dbm - +10dbm at peak time - before they moved (allegedly)! Any difficulty in this case was caused by the BC
modulation sidebands, not the BC transmitter's phase noise.
The AGC applied to the J309 comes from two sources via the op amp U3A.
One source is the HAGC detector D29/ D30, which receives its input from the 15
kHz IF amps U15A and U15B. I do wonder whether or not this HAGC detector introduces any IMD into the 15 kHz IF, being as it is not isolated in any
Control from the second source appears on the line VIF GAIN 1, which drives the op amp U3B which in turn drives the op amp U3A
The output of the op amp U3A acts on both the J309 and the diodes D21/ D22 to
control gain. In this scheme the gain of the J309 is changed by reducing
its Drain current, i.e. by changing its operating point, which could increase any IMD generated by the J309 just at the time when one wants an
improvement. I have similar reservations about the IMD performance of D21/D22.
The problem with the SA612 is its very low Intercept, fine for some portable QRP kit if large signals are not encountered, but a very poor choice of mixer
for use in this type of receiver.
1) The LO feed to the receiver's second mixer and the transmitter's first
mixer, derived from a divide-by-six divider, are not buffered but are simply
2) The receiver's first mixer could generate noticeable IMD, because unlike in the H-Mode mixer the Sources of the FETs are not connected
directly to ground.
3) Are the cores used in the bandpass filters, and elsewhere, large enough so that the IMD that they contribute is of no concern? Cores of size T94 are
required in a H-Mode receiver.
4) How much 'blow-by' exists around the roofers?
Whether or not the Group Delay Variations are in fact partly responsible for
the "mush" can only be determined, I believe, by some comprehensive
of the filters - and their effect on pulsed signals. Nevertheless it does appear that the designers were only considering raw selectivity, perhaps for
reasons of cost. I also wonder about the quality of crystals used in the filters.
If an input signal to a filter consists of a series of pulses such as a keyed
CW signal, and the filter exhibits any delay, then both the rise time
and the decay time of each pulse as seen at the output of the filter will not be the same as that of the original input. The decay time in particular
is related to the Q of the filter and can be significant in narrow bandwidth filters.
* * In the case where a filter exhibits significant group delay variations within its passband and there are many close spaced keyed CW signals
entering the filter, then a situation can arise where the stronger signals entering the filter do not appear to be the stronger signals at the output
because they take longer to reach their maximum amplitude than their weaker companions. * *
When the signals are weak, noise compounds the problem because noise spikes can be viewed as signals.
Although the effects of group delay variations can be calculated, the maths becomes very tedious if many CW signals keyed at different speeds are
involved. Dynamic tests are a better option!
Chris, the test that you suggest would be suitable provided that the odd
order IMD products generated by the combiner used between the generators and
the K3 are well below the K3's noise floor, and the isolation between the combiner's input ports is adequate. Of course it would be useful if both the
amplitude and p.r.f. of the signal from each generator could be varied on an individual basis.
The problem as I see it is to identify which element or elements in the signal path is/are responsible for creating the "pileup mush". Could someone
tell me please whether or not the "pileup mush" becomes less of a problem, or ceases to exist, if a wider bandwidth roofer is used, e.g. a 1800 Hz 8
pole vs. a 400 Hz 8 pole, while keeping the DSP filter's bandwidth constant.
If the "mush" does disappear when the wider bandwidth roofer is used, then this would suggest that the group delay variations found by Gavin in the
narrow bandwidth roofers is the prime suspect. If the "mush" does not decrease, then something else in the signal path is creating the problem.
There is also the middle ground where the "mush" decreases but is still heard, which would suggest that both the narrow bandwidth roofer and the
"something else" together are creating the "mush" problem.
Not to be forgotten is the two crystal filter between the J309 IF amp and the second mixer. If it is tuned properly and uses matched crystals, then
its group delay variations should not be large enough to create "mush".
At the moment I do not suspect the DSP's ADC. However if the delays in the narrow bandwidth roofer are causing half cycles to be lost, then this would
most likely affect the ADC.
I note also that you have heard the "mush" not only when signals are weak, but also when signals are quite strong. This raises the spectre of improper
filter terminations, and whether or not the application of HAGC affects the output termination of the roofer. I suspect that it might, but I have not
run a SPICE simulation to confirm my suspicion.
Thank you for sending me a copy of your interesting e-mail to Gavin.
With regard to the impact of filter group delay on 'real world' contesting, and in particular the effect of group delay variations within a filter's passband and skirt regions, I would suggest that while the basic theory exists the problematic question is 'how does one apply the theory to determine the impact of measured group delay variations on 'real world' contesting?'
Perhaps it would be possible to get some idea of the impact if some clever person wrote a computer program, which required as inputs the spacing in terms of frequency between several signals, their relative amplitudes as seen at the *input* of a filter, the keying speed of each signal, and the group delay imposed by the filter on each keyed signal. The rse and decay times of each signal as seen at the filter's input and the characteristics of any noise would also be required as inputs of course.
Producing such a program appears to me to be a difficult nut to crack, and although tedious dynamic tests would seem to be a better option, there appears to be no evidence which suggests that dynamic tests have been performed in the 'Ham world'.
With regard to the use of filters with tight skirts, it would appear that the majority, if not all of the designers of amateur receivers who have employed Roofing filters with tight skirts, have ignored the effects of filter group delay variations - probably for reasons of filter cost. I believe that a better approach is to employ a roofing filter whose skirts are quite gentle, followed by an IF cascade and back end whose odd order dynamic range is very large e.g. some 40db better than that of the K3's IF. By introducing additional filtering further downstream, the overall selectivity can be made 'tight' and the overall group delay variations within the passband and skirt regions can be kept to a suitably low value.
You mentioned the Harris R-2368/URR. Many years ago when I worked for RCA International in Canada, we developed an ancestor (I believe) of this receiver for use in HF ISB point-to-point applications. This receiver empolyed mechanical filters for its roofing filter and its USB-LSB filters, whose group delay variations were acceptable - I cannot remember the actual figures.
Section 2 - these filter plots were received from Gavin GM0GAV following his tests on the roofing filters in his K3 S/No.2203.
Gavin is going to test the roofing filters from his second K3 a.s.a.p.
These plots did not display 100% correctly when converted to HTML, so please click here to download these as a PDF document (156kB)
Geoff GM4ESD made some comments about the results of Gavin's tests - click here to download these as a Word document (574kB)
I am no expert but here goes.
The delay is not the issue, it is the change vs frequency. In the ideal filter the green line should be horizontal. Different frequencies are delayed by different amounts. As you know a fast cw signal is quite wide as it increases with speed. It contains sidebands which should all come through the filter at the same time. If not the envelope is distorted which reduces intelligibility, in other words start and stop of that dot becomes a bit fuzzy.
How big is the effect on an Inrad 400, I don't know. But it could just be one of several issues along with DSP. Also Geoff is looking at the termination of the filters in Spice, that could be making the group delay variations worse. My plots were all done with a nice resistive 500ohm pad.
May be smart money is on having a Elecraft 5 pole 500hz for pileups rather than an inrad 400, or better still both. 73 Gav
sorry forgot to add this link http://www.ac0c.com/main/page_ft2k_roofing_filters_ft5000_300_hz_plots.html
Section 3 - e-mails from Dave GM4EVS
27th March 2011 - more information to follow when available ...