Design Article
Bench measurements under 110dBc 3rd order intermodulation distortion
Michael Steffes, Sr. Applications Manager, High Speed Signal Path, Intersil Corp.
11/12/2012 1:45 PM EST
Source summing
What does matter is to get the two closely spaced signals summed together without the two sources talking to each other through the power combiner with a possibly non-linear output impedance. Any slight output stage nonlinearity can actually create IM2/IM3 terms in the combined source signal. Some isolation can be achieved simply by padding down the inputs and/or using a power combiner with high port to port isolation. For the ultimate in source to source isolation, the test signal for ISL55210 buffered each generator output with a very high intercept (linear output impedance) RF amplifier (ref. 4) before the power combiner as shown in fig. 4.
Click on image to enlarge.
The MiniCircuits Hela-10D amplifier used here are +12V supply requiring 525mA quiescent current (6.3W/each) but provide 11dB gain with an OIP3 at a matched 50Ω load >48dBm. The ZFSC-2-4-S+ power combiner in fig. 4 quotes approximately 30dB port 1 to port 2 isolation and about 3dB insertion loss to the summed output pins. The Hela-10D amplifiers were modified with an output resistive pi attenuator to generate a 0dB gain stage from the inputs at the Hela-10D’s, to the combiner output pins. This helps to keep both the loading on the Hela-10’s ok as cables were connected/disconnected and allowed the source programmed power to be placed as a test input power level to the DUT.
Extensive lab testing on the combined output 2-tone signal failed to expose a 3rd order spurious (down to -120dBc) even to very high (4dBm) output test signals. Recall this summed output is intended to drive an amplifier stage, so these 4dBm single tones combine to a 2Vpp envelope and should be adequate for most measurements. If used for ADC testing, this very low IM3 source will need to be filtered by a narrow bandpass filter to limit the out of band spectral (noise and distortion) artifacts as the ADC will fold all of those into Nyquist. Normally, much lower test source levels are required for amplifier testing. The intended application span was approximately 10Mhz to 200Mhz with the measured frequency response for each of the two port inputs of fig. 4 shown in fig. 5. The unused input in each test needs a 50Ω termination for this measurement.
Click on image to enlarge.
Measuring the output SFDR for a 2 tone IM3 spurious
Assuming we can generate a very clean input signal to test a modern, low power, FDA like the ISL55210, we are still left with the issue of resolving what might be more than -110dBc separation from the test tones to the spurious. Most spectrum analyzers are limited to about -85 to -90dBc dynamic range. While there are a number of tricks that can be used to extend this measurement range, the one selected here essentially creates a super narrowband notch filter on one of the output test tones from the DUT. This allows the spectrum analyzer attenuation to be greatly reduced without creating internal IM3 terms that would mask the DUT’s performance. Backing out the attenuation to the spectrum analyzer mixer will of course create very poor single tone HD terms, but those are not of interest in this test.
Prior to applying the DUT output spectrum to a spectrum analyzer, the first step is to attenuate to where the 2-tone power levels will not create meaningful IM3 if passed back through another Hela-10D amplifier. With a 48dBm intercept in the Hela-10D, and a -120dBc measurement aspiration, this implies taking the DUT output powers down to approximately -12dBm on each tone at the output of the Hela-10D. This implies approximately -23dBm for each tone at the inputs using eq. 2 and solving for P0. This impedance isolated DUT output spectrum (that is being passed through another HELA-10D), is then applied to another power combiner structure identical to fig. 4 where the 2nd input is now a 3rd low phase noise lab signal generator.
This 3rd phase locked source can be combined with the DUT’s output spectrum as shown in the complete test structure of fig. 6, where here the fixed resistive pi attenuator is not used on the outputs of the HELA-10D’s in this output side power combiner to keep the spurious as high as possible. The output of the 2nd power combiner then goes to a spectrum analyzer with minimal internal attenuation. The remaining uncancelled test tone gives an indication of P0, while the spurs at +/-3Δf can be measured by zooming in on a small span, with a low RBW, and sweep averaging.
Click on image to enlarge.
Summary and added projection techniques to estimate extremely low IM3 stages
The measurements here have shown<-110dBc IM3 performance on one of the emerging very high SFDR, low power, signal chain solutions. These devices have been applied successfully to both ADC interface (ref. 5) and IF applications saving significant power where suitable. While not used here explicitly, there are a number of other techniques that allow measurements in one range to be projected to others. For VFA based amplifiers, reducing the loop gain with a resistor across the inputs can be used to artificially increase the distortion terms during test, then project them to what they would be with the resistor removed. In theory, the SFDR should increase by the dB delta in noise gain. Also, once a solid measurement is made, such as those above 150Mhz in fig. 2, if the amplifier has a 20dB/dec loop gain rolloff, the distortion terms can be projected to run on that slope up and down in frequency.
And finally, if the amplifier can be shown to have an intercept characteristic, (i.e., step the test tone powers by 1dB and the 3rd order intermodulation terms should change by 3dB, 2dB change in SFDR) getting that intercept at relatively high output power levels will allow the SFDR to be projected at lower, possibly un-measurable, levels.
References:
What does matter is to get the two closely spaced signals summed together without the two sources talking to each other through the power combiner with a possibly non-linear output impedance. Any slight output stage nonlinearity can actually create IM2/IM3 terms in the combined source signal. Some isolation can be achieved simply by padding down the inputs and/or using a power combiner with high port to port isolation. For the ultimate in source to source isolation, the test signal for ISL55210 buffered each generator output with a very high intercept (linear output impedance) RF amplifier (ref. 4) before the power combiner as shown in fig. 4.
Click on image to enlarge.
Figure 4. Source summing structure for <-120dBc IM3 test signals (power cables removed).
The MiniCircuits Hela-10D amplifier used here are +12V supply requiring 525mA quiescent current (6.3W/each) but provide 11dB gain with an OIP3 at a matched 50Ω load >48dBm. The ZFSC-2-4-S+ power combiner in fig. 4 quotes approximately 30dB port 1 to port 2 isolation and about 3dB insertion loss to the summed output pins. The Hela-10D amplifiers were modified with an output resistive pi attenuator to generate a 0dB gain stage from the inputs at the Hela-10D’s, to the combiner output pins. This helps to keep both the loading on the Hela-10’s ok as cables were connected/disconnected and allowed the source programmed power to be placed as a test input power level to the DUT.
Extensive lab testing on the combined output 2-tone signal failed to expose a 3rd order spurious (down to -120dBc) even to very high (4dBm) output test signals. Recall this summed output is intended to drive an amplifier stage, so these 4dBm single tones combine to a 2Vpp envelope and should be adequate for most measurements. If used for ADC testing, this very low IM3 source will need to be filtered by a narrow bandpass filter to limit the out of band spectral (noise and distortion) artifacts as the ADC will fold all of those into Nyquist. Normally, much lower test source levels are required for amplifier testing. The intended application span was approximately 10Mhz to 200Mhz with the measured frequency response for each of the two port inputs of fig. 4 shown in fig. 5. The unused input in each test needs a 50Ω termination for this measurement.
Click on image to enlarge.
Figure 5. Source summing frequency response.
Measuring the output SFDR for a 2 tone IM3 spurious
Assuming we can generate a very clean input signal to test a modern, low power, FDA like the ISL55210, we are still left with the issue of resolving what might be more than -110dBc separation from the test tones to the spurious. Most spectrum analyzers are limited to about -85 to -90dBc dynamic range. While there are a number of tricks that can be used to extend this measurement range, the one selected here essentially creates a super narrowband notch filter on one of the output test tones from the DUT. This allows the spectrum analyzer attenuation to be greatly reduced without creating internal IM3 terms that would mask the DUT’s performance. Backing out the attenuation to the spectrum analyzer mixer will of course create very poor single tone HD terms, but those are not of interest in this test.
Prior to applying the DUT output spectrum to a spectrum analyzer, the first step is to attenuate to where the 2-tone power levels will not create meaningful IM3 if passed back through another Hela-10D amplifier. With a 48dBm intercept in the Hela-10D, and a -120dBc measurement aspiration, this implies taking the DUT output powers down to approximately -12dBm on each tone at the output of the Hela-10D. This implies approximately -23dBm for each tone at the inputs using eq. 2 and solving for P0. This impedance isolated DUT output spectrum (that is being passed through another HELA-10D), is then applied to another power combiner structure identical to fig. 4 where the 2nd input is now a 3rd low phase noise lab signal generator.
This 3rd phase locked source can be combined with the DUT’s output spectrum as shown in the complete test structure of fig. 6, where here the fixed resistive pi attenuator is not used on the outputs of the HELA-10D’s in this output side power combiner to keep the spurious as high as possible. The output of the 2nd power combiner then goes to a spectrum analyzer with minimal internal attenuation. The remaining uncancelled test tone gives an indication of P0, while the spurs at +/-3Δf can be measured by zooming in on a small span, with a low RBW, and sweep averaging.
Click on image to enlarge.
Figure 6. Complete source and output tone nulling structure for <-110dBc IM3 measurements
Using
the 3rd generator with phase tuning, it is then possible to set the
amplitude and phase of a cancelling signal in this 2nd power combiner to
null out one of DUT output tones. Actual results showed about a -50dBc
selective attenuation on one of the tones using the HP8644 source. This
proved adequate to extract the measured results down to the -120dBc
level of fig. 2. Since all of these sources and spectrum analyzers are
available under GPIB control, this test methodology could be expedited
by programming it under Labview control. Summary and added projection techniques to estimate extremely low IM3 stages
The measurements here have shown<-110dBc IM3 performance on one of the emerging very high SFDR, low power, signal chain solutions. These devices have been applied successfully to both ADC interface (ref. 5) and IF applications saving significant power where suitable. While not used here explicitly, there are a number of other techniques that allow measurements in one range to be projected to others. For VFA based amplifiers, reducing the loop gain with a resistor across the inputs can be used to artificially increase the distortion terms during test, then project them to what they would be with the resistor removed. In theory, the SFDR should increase by the dB delta in noise gain. Also, once a solid measurement is made, such as those above 150Mhz in fig. 2, if the amplifier has a 20dB/dec loop gain rolloff, the distortion terms can be projected to run on that slope up and down in frequency.
And finally, if the amplifier can be shown to have an intercept characteristic, (i.e., step the test tone powers by 1dB and the 3rd order intermodulation terms should change by 3dB, 2dB change in SFDR) getting that intercept at relatively high output power levels will allow the SFDR to be projected at lower, possibly un-measurable, levels.
References:
|
|
|
|
|
Navigate to related information