Intermodulation testing of high performance receivers

Some thoughts on test methods and possible pitfalls

Essentially the problem is two-fold - firstly to generate a two-tone signal with as little distortion as possible, and secondly to establish that the measured results apply to the receiver under test and not to a combination of receiver and test equipment.

Addressing two-tone signal generation first, a good hybrid combiner (eg Mini-Circuits PSC2-1) is blameless at signal levels up to 0dBm. Its IP3 is in the region of +50 to +60dBm (the exact value depends on frequency and port impedances) so it will not give problems until receivers get up to +45dBm IP3 points. The biggest culprit producing IMD is the ALC loops in the output stages of the signal generators. Each generator detects some of the signal from the other generator as well as its own and the resultant beat frequency is amplified and applied to the output signal as AM. Sadly the de-facto standard, low-noise generator, the HP8640 is very prone to this and has an ALC bandwidth extending to several hundred kHz. (From what other people have reported most HP generators are poor in this respect, with R&S and Marconi Instruments products fairing better. Sadly intermodulation resistance to reverse power is not a specified parameter.) The amount of IMD produced by the generator ALC is dependent on the generator to generator coupling which in turn is dependent on hybrid coupler isolation and, more importantly, DUT input impedance (power is reflected if VSWR is not unity). I think that differences in antenna input VSWR is one of the causes of measurement differences between different high performance radios on suspect test equipment. (This implies that the argument "Radio X measures at IP3 = +20dBm, so a measurement on radio Y of +15dBm is correct." is not valid).

I have used three methods to improve the IMD performance of the two-tone signal source

1) Replace one or both signal sources with crystal oscillators followed by grounded gate FET buffer amplifiers giving +10dBm output, which is then attenuated to the desired level.

2) Add a high gain amplifier and then an attenuator between the signal generator and the combiner unit. The amplifier should be operated well within its linear range (maybe >10dB below rated output). I have successfully used +40dB amplifiers followed by 20dB of attenuation.

3) Modify the ALC system of the signal generator sources.

Any one of the above should improve the quality of signal fed into the receiver. With modified HP8640s I can achieve IM products more than 80dB below two signals each at +6dBm (verified with spectrum analyser). The crystal oscillators give a few dB better on a good day. Taking -90dB as a realistic target for the purity of the two-tone source means that any tests should aim to produce intermodulation in the receiver under test at least 10dB above this level for confident results. With signal injection to the receiver at 0dBm this sets a test equipment limitation of performance to an IP3 at +40dBm - good enough at the moment.

Now to the actual testing. A quick thought will confirm that the higher the signal level the receiver is tested at, the less critical is the quality of the source. Testing at the noise floor level of the receiver will place unrealistic demands on the signal source if the receiver performance is good. I use receiver S-Meter readings to measure IM products at higher levels, so a test procedure goes something like this :-

1) Connect signal sources to combiner and adjust to give two signals each at 0dBm at the combiner output.

2) Feed the two-tone signal through a 1dB-step attenuator to the receiver input and tune the receiver to an IM product

3) Adjust attenuation until an S-Meter reading is obtained corresponding to a signal at about -80dBm (this should be about S9 on most receivers).

4) Measure the IM product signal level accurately by substituting a single signal at the receiver input and adjusting for a similar S-Meter reading.

5) The test can be repeated several times for slightly different two-tone signal levels.

IP3 can be calculated from the measured results as :-

<IP3> = ( <two-tone input level> - <single signal input level> ) / 2 + <two-tone input level>
(all levels in dBm)

Dynamic range can then be calculated by reference to the receiver's noise floor :-

<IFDR> = ( <IP3> - <noise floor> ) * 2 / 3

It is worth being aware of the pitfalls in this test method. To date I have only thought of one, which is that some receivers may start to invoke RF gain reduction from AGC action at the levels tested. This will certainly show up as an anomaly in the results (see later), but so far I have not met a "good" receiver that starts this effect below signals of S9+10dB. The AR 7030 does not add RF attenuation until S9+40dB.

The most important facility that this test allows is verification. Because a range of input levels can be used the connection between two-tone input level and IM product level can be evaluated and compared with the theoretical 1dB / 3dB relationship. If the test results can match this over several dB of input levels then one can be fairly sure that the IM products measured are developed in the DUT.

As an example I have obtained the following results from a sample of the AOR AR7030 receiver at 100 kHz tone spacing (noise floor at -123dBm) :-

Two-tone
input level
Equivalent IMP
input level
Calculated IP3 Calculated IFDR
0dBm -68dBm +34dBm 104.7dB
-1dBm -72dBm +34.5dBm 105dB
-2dBm -74dBm +34dBm 104.7dB
-3dBm -78dBm +34.5dBm 105dB
-4dBm -80dBm +34dBm 104.7dB
-5dBm -84dBm +34.5dBm 105dB
-6dBm -87dBm +34.5dBm 105dB

IM products rising at less than the 1dB / 3dB rate would indicate reception of a spurious response (rather than an IM product) or reception of a distortion product from the test equipment. Products rising at higher rates or rising erratically merit further investigation, often plotting a graph of input level vs product level is useful to attempt a straight line (of slope 3) through a scatter of points. Such effects seem to occur when several stages in the receiver generate IM products at more or less the same level, or when AGC action invokes a PIN diode RF attenuator (which usually increases IMPs). Predicting an IP3 value from erratic results has to be a matter of judgement!

It is interesting to note that when similar levels of IM products are produced in the receiver and the test setup, or in several different stages in the receiver, that the signals may add (producing lower IP3 results) or cancel (producing higher IP3 s). Cancellation usually only occurs over a narrow frequency range (maybe only at one receiver frequency or one particular tone separation) but can produce bizarre results 10 or 20dB in excess of proper IP3 values.

Intermodulation

Intermodulation :- I had thought, and indeed hoped, that the receiver intermodulation measurement debate had been finally laid to rest but it seems that Radio Netherlands have exhumed the corpse ready for a second post-mortem. So, donning appropriate rubber gloves and apron, here goes....

The Idea of using third-order intercept point (IP3) and Intermodulation-free dynamic range (IFDR) as performance specifiers for a receiver is that different receivers can be compared like for like with regard to their strong signal handling capabilities. Given that test conditions are similar (frequency, signal separation, receiver filter bandwidth etc) then figures should be directly comparable. RN now introduce a further complication - how the measurement was done and indeed publish four different results in their latest review of the AR7030 (December 1996). It is not as if their results were broadly similar - they obtain IP3 figures from +1dBm to +33dBm, a difference of more than three orders of magnitude, in receiver terms ranging from a mid-price portable to a very expensive professional rack-mount receiver. They cover this anomaly by saying that "These values ... show that when you compare published receiver intercept points it is essential that you know the level of the [test] signals otherwise the IP3 figure is meaningless."

Is the IP3 figure such a difficult animal to quantise, depending on external conditions ? Well, no, of course it isn't, and it's been used as a reliable indicator of circuit performance for many years in the radio engineering industry. So who is correct then ? Well I think that I am, and I can prove it (but then I would say that wouldn't I ?).

Mathematical interlude :-

On the basis that every equation will halve the readership of an article (S. W. Hawking) this is kept as short as possible.

The transfer characteristic of a circuit, which describes its output in terms of its input, can be represented as a polynomial equation. A perfectly linear circuit would have zero coefficients for all polynomial terms above first order - it would produce no distortion. A real circuit has non-zero higher order terms which means that the signal passing through it is distorted to some degree. In considering the third order distortions the critical part of the equation is the cubic term. This means that the level of distortion is proportional to the cube of the input signal level, in other words if the input increases by 10dB then the third order products at the output would increase by 30dB.

This input / output relationship is fundamental to the concept of intercept point (which is a theoretical point at which the intermodulation products are at the same level as the signals generating them).

Proof 1 (Confidence in the test equipment) :-

Any intermodulation measurement requires that several (typically two) signals are injected into the unit under test (UUT) and then the levels of any resultant intermodulation products produced in the UUT are measured. The IMD levels are typically referred back to their equivalent input levels for calculation of IP3 and IFDR. It is important to realise that possible sources of IM products extend outside the UUT into the signal sources and measurement equipment as well. For the purposes of this proof I will divide the test setup into three parts:-

Part 1 is the signal source, consisting of oscillators or signal generators and a combining unit.

Part 2 is a variable attenuator (to control the level of signals fed into the receiver).

Part 3 is the receiver under test, with any associated audio analysis equipment.

It is assumed that Part 2 (the attenuator) does not generate any intermodulation products, i.e. it is blameless. Also although the audio analysis equipment is lumped into Part 3 it is not implicated in IMD production.

Both Part 1 and Part 3 are capable of producing intermodulation products. RN claim that their crystal oscillator signal sources are "combined in such away (sic) that they do not influence each other" but offer no evidence to support this statement (note 1).

The signal levels in Part 1 (before the attenuator) are essentially constant, varying only slightly as the attenuator is changed with any mismatches into Part 3. Any IMD produced in part 1 should therefore be at a more or less constant level.

The signal levels in Part 3 vary with the attenuator setting. Any IMD produced in Part 3 should therefore follow the cubic rate law (as above) and change three times faster than the signal level.

Here are the figures from the RN tests :-

Measurement
level
Two-signal
input
Equiv IMD
level
IFDR IP3
1 -40dBm -122dBm 82dB +1dBm
2 -29dBm -107dBm 88dB +10dBm
3 -10dBm -87dBm 100dB +28dBm
4 0dBm -66dBm 103dB +33dBm

Let's first assume that the signal sources are blameless, and all of the IMD is generated in the receiver :-

Measurement
Two-signal
level
Equiv IMP
input level
IMD level
wrt signal
Expected
IMD level
Error
1 -40dBm -122dBm -82dB -146dB 64dB
2 -29dBm -107dBm -78dB -124dB 46dB
3 -10dBm -87dBm -77dB -86dB 9dB
4 0dBm -66dBm -66dB -66dB (note 2)

Let's now assume that the receiver is blameless, and all of the IMD is generated in the signal sources :-

Measurement
level
Two-signal
level
Equiv IMP
input level
IMD level
wrt signal
Expected
IMD level
Error
1 -40dBm -122dBm -82dB -82dB (note 3)
2 -29dBm -107dBm -78dB -82dB 4dB
3 -10dBm -87dBm -77dB -82dB 5dB
4 0dBm -66dBm -66dB -82dB 16dB

A quick look at the last column in the above two tables indicates much lower error values in the second case. This means that the second hypothesis more closely fits the data - indeed if we disregard the fourth measurement (because at 0dBm input levels we know that the receiver is NOT blameless) then the error values are more or less within an expected +/- 2dB measurement error for this type of test (note 4).

The conclusion of this proof is that the RN test source generates intermodulation products at 80dB (+/- 2dB) below the level of the wanted signals. Using their test method with this source will misrepresent the performance of any radio with an IFDR greater than 78dB.

Proof 2 (How does this work then ?) :-

The RN intermodulation results in the first table (above) indicate a receiver that improves (produces less distortion) as the signal levels get larger. If these measurements are to be believed then to explain this behaviour requires a circuit in the receiver whose distortion firstly increases with increasing signal level, but then decreases as signal levels increase further. It is quite a challenge to design a circuit that behaves like this, and although I could see that a class AB push-pull amplifier might have this characteristic, the class A stages in a receiver would be hard-pushed to mimic it.

This second proof is not as conclusive as the first (since it invokes "not imaginable" as reason for non-existence), indeed some may query if it is a proof at all, but it does avoid a lot of figures, and some may find it a convincing reason to doubt what RN have published.

Notes

1) It is fallacious to think that a 6dB pad (fixed attenuator) ensures immunity against impedance mismatch. In fact any adverse effect is only reduced by 6dB because although the reflected power is attenuated by 12dB the source power has to be increased by 6dB to overcome the attenuator.

2) The last measurement is the least taxing for the sources so this is used as the 0dB error reference point.

3) The first measurement is the least taxing for the receiver so this is used as the 0dB error reference point.

4) Errors mainly come from receiver s-meter resolution, attenuator accuracy and impedance mis-matches.

Intermodulation News - more IP3 thoughts and tests

The AR7030 receiver has proved to be an excellent platform for investigation of 3rd order intermodulation distortion. Firstly its performance is good enough to tax most of the commonly available test equipment and secondly its signal path has sufficiently few stages so that intermodulation from each one can be measured and isolated.

Recent tests on the 7030 by the ARRL have shown up some strange intermodulation effects which prompted me to re-visit all of the IMD testing with an improved test setup and provide an explanation for the observed results. These particular anomalies are discussed in the last part of this article which is an unashamedly technical (but non mathematical) look at intermodulation within HF receivers, methods of intermodulation testing and components that generate IMD.

This article concentrates on third order intermodulation although its comments and conclusions are equally appropriate to both even and odd-order distortions. Odd-order intermodulation (3rd order usually being the most prominent) is normally used as an assessment of a receiver's dynamic range because there is very little that can be done external to the receiver to improve it. Even-order distortions can generally be reduced by filtering the receiver's input (pre-selection) because at least one of the signals causing interfering intermodulation products is well separated in frequency from the wanted signal.

Intermodulation test setup.

In a standard test situation intermodulation products are generated between two equal level signals separated by a specified frequency. Typically two signal sources (signal generators) are used with their outputs combined into the antenna input of the receiver. Additionally a single signal (at the frequency of the expected intermodulation product) is usually needed to calibrate the receiver so the level of the IMD product can be determined.

The main difficulty in generating and combining the two signals for the test is to prevent intermodulation within the test equipment, or at least ensure that it is at a level below the expected IMD in the receiver under test. Because the 7030 is very demanding in this respect, I have previously suggested that testing should be done at a sufficiently high level to ensure that receiver IMD will swamp any input from the test equipment. I still think that this is a valid approach, but wanted to improve my own testing capability to include assessment of receiver IMD at low signal levels.

There are three main sources of intermodulation within the test setup :-

1) Distortion in the combining network. A simple resistive combiner will not generate distortion but will also not provide much isolation between the two signal sources which, in fact, is a bigger problem. A hybrid combiner is more satisfactory, but contains broadband transformers which can suffer from core saturation. In practice a Mini-Circuits PSC2-1 has an IP3 in the region of +50 to +60dBm (the exact value depends on frequency and port impedances) so it will not give problems until receivers get up to +45dBm IP3 points. It is, however, worth operating the combiner at the lowest possible signal levels, so there should be no signal attenuation between the combiner and the receiver's antenna input.

2) Intermodulation in the output amplifiers of the signal sources (due to cross-coupling of the two signals). The higher the power of the amplifier the better, so signal generators capable of +19 or +20dBm output are recommended. An additional linear amplifier after the signal generator can work well, but the common "single device" type of broadband amplifiers do not seem to be very good at rejecting signal at the output, and there can be additional problems of increased broadband noise. In any case the test setup should try to minimise cross-coupling (see later).

3) Unwanted modulation of the signals by the level control system within the signal generators. Obviously this only applies to test setups using levelled signal generators as sources, but beware - many crystal oscillators have ALC systems to ensure low phase noise and these are just as vulnerable. The effect is due to the signal generator striving to maintain a constant output level by rectifying its output signal and using a feedback system to drive an amplitude modulator. If a second signal is applied to the output, then the detector will produce a signal at the "beat" frequency of the two signals (ie at the difference frequency) and this signal will modulate the generators output. The frequency of the modulation sidebands will neatly coincide with the expected IMD product frequency that we are looking for in the receiver under test ! The reason I have separated this effect from intermodulation in (2) above is that it is linear wrt the second signal level fed into the generator's output whereas an intermodulation effect would change at a third order rate. It is therefore difficult to isolate this effect by changing signal levels and attenuations within the test setup. I have only managed to quantify it by taking measurements within the ALC loop of the signal generator.

The distortion from all of the above effects can be minimised by ensuring that there is as much attenuation as possible between each signal source and the combiner. Many signal generators do not maximise output attenuation, preferring to run their output amplifiers at lower than maximum levels (for better harmonic performance) so it is usually preferable to run the generators at maximum output and use an external 1dB step attenuator to control the level into the combiner. In the case of the HP8640 generators that I use this gives between 20 and 40dB more generator-generator isolation than simply connecting the combiner directly.

The ALC modulation problem is severe in the HP8640 (it certainly affects IMD results in receivers with IP3 above +15dBm) but can be virtually eliminated by a simple modification to restrict the bandwidth of the ALC system to about 1Hz. This mod removes the AM capability of the generator, so it has to be switchable (I can provide details if anyone wants).

As well as generating the intermodulation test signal the test setup must be able to establish the level of intermodulation distortion that the receiver produces. It is normal to treat the receiver as a "black box" with an antenna input and an audio output and use only these signals (rather than inserting probes into the receiver circuit to inject or monitor at intermediate points). In order to use the receiver's audio output to assess its IMD performance it must be used in a "linear" reception mode where the output reflects what happens at the input. Typically for an HF receiver this would be one of the SSB modes or CW - the intermodulation products can be resolved at a frequency near to the centre of the receiver's passband, say 1kHz.

There are three ways of determining the IMD product level in a "black box" receiver. All have limitations, and often a combination of methods is required to produce comprehensive and confident results :-

1) Calibrated S-Meter.
The audio output of the receiver is not used other than as an aural tuning aid, instead the level of the IMD product is noted on the receiver's S-meter display and then a single signal is fed into the receiver to give the same S-meter reading. This method is good provided that the resolution of the S-meter is good - the absolute calibration of the meter does not matter - and favours testing the receiver at higher signal levels where the IMD products are well clear of the noise floor. In fact it is often the only way to determine the higher levels of IMD products. Beware that some receivers (not the AR7030) start to reduce the gain of the RF stages when the S-meter gets to the S9 or S9+10dB region - this will produce erroneously high results for IP3. Testing in the S3 to S7 region is generally OK.

2) Audio output S/N ratio.
The audio output is tuned to exactly 1kHz and analysed with a signal / noise ratio meter or SINAD meter. Accurate tuning and low drift in the receiver and signal sources is a requirement but is not normally a problem with synthesised equipment. It is possible to estimate S/N with an audio power meter but this must be done very carefully because readings can be compromised by receiver AGC action - switch the AGC off if possible but then beware of limiting in the IF or audio stages. The S/N ratio measurement is good for signals at or up to 20dB above the noise floor of the receiver and give a reading relative to this noise floor. The big problem is that the noise floor is often raised during the IMD testing either by reciprocal mixing in the receiver or by phase noise from the test generators and this gives an impression that IMD levels are lower than they actually are. The noise increase can be largely compensated for by continuing to feed the test signal closest in frequency to the IMD product whilst introducing the calibrating signal (at a level to give the same S/N ratio as the IMD product). This assumes that the majority of the noise or reciprocal mixing comes from the test signal nearest to the receiver's tuned frequency. This test method is also very susceptible to spurious responses in the receiver if using a S/N meter to analyse the output. Any heterodynes thus produced will be considered as part of the noise - so always listen to the output and if necessary change the test frequencies slightly to avoid spurii (it is often a problem with 50kHz spaced signals - try 45 or 55kHz instead).

3) Audio signal level.
Using an AF spectrum analyser or selective voltmeter (wave analyser) the level of the IMD product can be measured at the audio output irrespective of the noise level on the audio. Indeed with a narrow filter in the analyser (say 10Hz) it is quite possible to measure levels that are well below the noise floor of the receiver. This technique can give good results over quite a range of signal levels, but initially the receiver needs calibrating with a single signal and the maximum measurable level determined. This will either be the onset of AGC gain reduction or, if AGC is switched off, the limiting level in the IF or AF stages.

The purpose of the above ramblings and my equipment modifications is to be able to investigate the levels of IMD products in a 7030 receiver over as wide a range of signal levels as possible and to verify that the input / output ratio follows the expected 1:3 relationship. If this is indeed the case then the computed value of IP3 should be a constant for all input levels.

The 7030 was tested with signals at 14.020 and 14.070MHz (50kHz spacing) in USB mode with the standard 2.2kHz filter selected. The IMD product was resolved at 1kHz with the receiver tuned to 13.069MHz. For low level inputs (up to -8dBm) the AGC was turned off and the audio signal level measured using a spectrum analyser with a 10Hz bandwidth. For higher signal levels (-6 to 0dBm) the AGC was turned on and the S-meter readings used. With the AGC off the 7030 is linear with antenna inputs up to -90dBm.

Two-signal input
level (each signal)
Equivalent IMD
product level
(ref to signal)
Calculated IP3
0dBm -66dBm +33dBm
-2dBm -73dBm +33.5dBm
-4dBm -78dBm +33dBm
-6dBm -84dBm +33dBm
-8dBm -89.5dBm +32.7dBm
-10dBm -94.8dBm +32.4dBm
-12dBm -100.5dBm +32.3dBm
-15dBm -109.7dBm +32.4dBm
-20dBm -124.6dBm +32.3dBm

These test were done with the RF preamplifier switched off, but the test results repeat almost exactly with the preamplifier turned on and the input signal level reduced by 10dB. The results show that the 1:3 input / output ratio is accurately maintained over a 60dB range of output from levels below the noise floor to over S9.

Intermodulation in HF receivers.

From a radio designer's point of view every receiver front-end is a trade off between sensitivity and dynamic range - the trick being to balance the gain (and hence signal levels) in each stage with the bandwidth of signals that it is subject to. If filtering is done before enough gain stages then sensitivity suffers. If too much gain is applied before filtering then there is a greater possibility of overload and intermodulation. In a typical dual conversion HF receiver there are two stages of filtering that affect the dynamic range (and IMD performance) of the receiver discounting, for the moment, any RF stage filtering.

Normally the dual conversion receiver up-converts the RF signal to the first IF which is typically between 40 and 80MHz. There the signal is filtered, usually with a crystal filter of 15 to 20kHz bandwidth, and amplified before being down-converted to the second IF which is normally below 2MHz, 455kHz being common. The receiver's main selective filtering is done in this second IF stage with bandwidths depending on reception mode - 2.2kHz is normal for SSB reception. The dual conversion arrangement is popular because it works well, providing good to excellent image rejection and good filter shape factors without using excessively expensive components. Tuning is also straightforward requiring only one tuneable local oscillator (for the up-conversion to the first IF).

With a few exceptions most HF receivers have no selective RF stages (after all the up-conversion has diminished the image rejection circuit to a simple RF low-pass filter cutting off below the IF frequency) or include only switched sub-octave filtering which helps with even-order intermodulation but does nothing for odd-order effects. The signal losses through the filters are a problem in the sensitivity stakes anyway. This means that all of the receiver's circuit up to the first IF filter (called the roofing filter) sees the whole of, or at least a large part of, the RF signals coming from the antenna. After that filter the circuit only sees signals within about 20kHz of the tuned frequency up to the selective filter which then eliminates everything except the required audio bandwidth.

By choosing IMD test frequency separations appropriately it is possible to establish the performance of each section of the receiver. Separations greater than 50kHz mean that the test signals will be blocked by the roofing filter and so this tests the RF stages, the up-conversion mixer and any IF amplifier before the roofing filter. It is common for there to be little improvement in IP3 figures once the 50kHz separation is exceeded - this simply indicates the lack of any RF selectivity in a receiver. Testing at a 4 or 5kHz separation will allow both IMD test signals through the roofing filter virtually un-attenuated so the next stage of circuit up to the selective filter is tested. This includes the first IF amplifier, the down-conversion mixer and the first stage of the second IF amplifier. Because these circuits have some gain (often a significant amount) the IP3 figure will be lower than for the 50kHz test. At test signal separations of a few hundred Hz the whole of the receiver is tested (assuming a 2kHz bandwidth) though the results of this test tends to reflect on the receiver's audio quality rather than its dynamic range.

In my opinion many receivers have too much gain before the selective filter, and the down-conversion mixer can cause severe IMD products when listening to signals close to strong stations. I have tested some receivers that give good +15 to +20dBm IP3 performance at 50kHz spacing but very poor results in the region of -30dBm at 5kHz spacing. Needless to say the 7030 is not designed like that, and the IF amplifiers and down-conversion mixer will return an IP3 of +12dBm. The price paid, of course, is in sensitivity, but even with 10dB of RF pre-amplification the close in IP3 should be better than 0dBm.

So what happens when the IMD test signal separation falls between the 5kHz and 50kHz values? Well I would expect a gradual transition of values as more of the test signal passes through the roofing filter with narrower frequency separations. And up to a point that is what happens, but some strange happenings can be observed along the way…

Intermodulation in passive components.

In April the ARRL carried out some IMD tests on the 7030 as part of their receiver review. They tested the receiver at 20kHz frequency spacing with the RF pre-amplifier switched on and obtained an IP3 result at around +12dBm which is some 12dB below the expected value. They had tested the receiver with an IMD product level at the noise floor, but because of the discrepancy with AOR's measurements continued the test at different signal levels and produced a graph of signal level vs. IMD product level. This showed an unusual characteristic; the IMD product level rose quickly from the noise floor by about 5dB for a 1dB input change and then stayed more or less constant for a further 10dB change. At higher input signal levels it then started to follow the expected 1:3 slope which corresponded to an IP3 of +25dBm (more or less to specification). I repeated their tests and found similar strange effects although I was not able to duplicate the results exactly.

Detailed investigation showed that the effect was very frequency dependent (it did not occur at all at frequency separations above 25kHz) and was often inconsistent, giving different results on different test runs. A gradual elimination of components that may cause the effect pointed to the roofing filter being the guilty party and indeed slight warming or cooling of that part caused significant changes in the IMD level virtually confirming it as the culprit.

That IMD should occur in a crystal filter is not surprising. The design of the RF stages in the 7030 has already been changed after initial production because of IMD in the 1.7MHz high-pass / low-pass filter. In this case both inductors and capacitors were responsible for IMD products near the filter's cut-off frequency (where voltages and currents in the filter tend to be highest) and the problem was solved by specifying larger inductors and a different dielectric for the capacitors. The point is though that the IMD in these passive components behaved in the expected manner - the product level rose at 3 times the test signal level.

Whatever is happening in the roofing filter cannot be described with a standard polynomial transfer function - indeed the input / output characteristic must actually change with the applied signal level and as such its IMD performance cannot be characterised with an IP3 figure. Since the filter is a mechanical device - energy is transferred from the input to the output by way of physical motion - the best mechanism I can think of for the observed behaviour is some kind of rattle. If anyone with more knowledge of filter technology wants to offer explanations I am happy to listen.

To summarise the observations - the roofing filter produces IMD products when one or both of the test signals are in the transition band of the filter (between pass-band and stop-band) and the level of these products does not obey the expected 1:3 input / output level ratio. The roofing filter in the 7030 is a four-pole fundamental mode crystal filter with 15kHz bandwidth at 45MHz made up of two monolithic dipoles. Its input and output impedances are matched to 500 ohms and it is driven from a heavily damped tank circuit. A 0dBm signal into the receiver's antenna input at the tuned frequency produces 510mV of IF signal at the input to the filter and 380mV at its output.

Conclusions.

The work on signal generator isolation has now produced an IMD test setup that is capable of measuring the 7030's performance down to and below the receiver's noise floor.

The resultant tests show that for signal separations of 30kHz or more the receiver behaves in a "textbook" fashion with consistent IP3 results irrespective of the testing signal level or the test method used.

There are some as yet unexplained anomalies in the IMD performance of the roofing filter.

© Copyright AOR UK LTD 1996 - 2008.

Get Back To The Top?