A COMPARISON OF 3 DIFFERENT NOISE SOURCES ON 1296 MHz.
(July 20 2013)

Background.

Noise figures are conveniently measured by use of dedicated instruments. A Noise Figure Analyzer (NFA) that is used with a calibrated Noise Source These instruments typically provide an accuracy in the order of 0.5 dB.

For 1296 where the sky temperature is about 3 K a measurement accuracy of 0.5 dB is totally unacceptable. That error is one order of magnitude larger than the sky temperature. Good modern amateur amplifiers have a noise figure of about 0.15 dB which corresponds to about 10 K noise temperature.

When building and comparing such amplifiers we need an improvement of one order of magnitude on the instrument accuracy.

By use of the cold sky and a room temperature reference Sergey Zhutyaev, RW3BP has estimated the NF of 1296 MHz amplifiers. I have myself used ice and boiling water to measure NF on 144 MHz. In both cases our completely independent measurements agree far better than expected with the noise figures obtained from standard instruments. It could be a coincidence, but it is rather unlikely.

The comparisons with NFA measurements are done in a way that eliminates many errors. If we assume the error elimination is 100% complete there is still the calibration error of the noise source. Such calibration errors are typically in the order of 0.2 dB which means that an agreement within 0.02 dB is a fairly unlikely coincidence. Two independent such coincidences represents a highly unlikely event and it seems more likely that the error limits are a bit to conservative.

In 2012 Pete, WA2ODO, was inapproprietly scorned on MOON-NET for having published NF values with two decimals, something that would not have been appropriate if the uncertainty really were as bad as claimed by the NFA manufacturer (Agilent).

This page, a cooperation between Mart, SM0ERR and myself, Leif, SM5BSZ is an attempt to shed some more light on the situation.

The NFA.

The NFA is an instrument that measures Y-factors. It may introduce errors of several kinds, IF amplifiers may saturate and analog power detectors are not quite perfect. Those things are problems in older instruments like the Agilent 8970A and 8970B. Those errors can be avoided but by use of a modern instrument, the Agilent N8973A one should get the proper Y-factors directly.

The amplifiers.

Three different amplifiers were measured.


Measurements with 8970A and circulators.

The same system that was used in Orebro at the EME2013 meeting was used to measure noise figures. The results are displayed in table 1. 

Unit     Temp      NF        NF26
          (C)     (dB)       (dB)       
SM5BSZ   27.7     0.1506     0.1474
SM0ERR   30.0     0.2050     0.1974
AD6IW    30.2     0.2754     0.2674
Table 1. Noise figures of the three amplifiers. The results are normalized to a box temperature of 26 C by use of a temperature coefficient of 0.13 K(Te)/C(room) or 0.0019 dB(NF)/C(room).
The data in table 1 is from the average of about 300 averages of 64 averages from the 8970A and the statistical errors are very small. About 4 times smaller than the statistical errors at the Orebro meeting.

Measurements with the N4000A noise head.

The serial number is MY44420135. It was calibrated May 20 2013 with this certificate.

Noise sources do not have the same impedance in the on state as in the off state. This is a source of an error, but by inserting a quarter-wave cable one can change the sign of the error and by using the average of the measurement with and without a quarter-wave cable one can eliminate this particular error.

The N4000A has an N connector while the tested amplifiers all have SMA female connectors. The N to SMA adapter was assumed to add 0.63 K, a number that comes from the Orebro meeting. The quarter-wave cable, DUTA was assumed to have a loss of 0.036 dB in accordance with the findings here. For the measurements without a quarter-wave cable the loss before the DUT was set to 0.009 dB and for the measurements with a quarter-wave cable the loss before the DUT was set to 0.045 dB. The loss after the DUT was set to 0.270 dB which was the loss of the cable plus the adapters used between the DUT and the NFA. The temperature of losses before and after the DUT was set to the room temperature.

Many measurements with 256 averages were made in each case. The raw data and evaluation of the AD6IW amplifier is presented in figure 1.

Figure 1. The L LNA measured with the N4000A noise source.

The temperature on the L LNA was 31.1 to 30.9 C during the measurements in figure 1 so the result at 26 C should be 0.2584 dB. The quarter-wave cable causes a change of 0.0047 dB on the NF when using the N4000 noise source.

Figure 2 shows the raw data and evaluation of the SM0ERR amplifier with the N4000A noise head.

Figure 2. The SM0ERR LNA measured with the N4000A noise source.

The temperature on the SM0ERR amplifier was 29.6 C during the measurements in figure 2 so the result at 26 C should be 0.1986 dB. The quarter-wave cable causes a change of 0.042 dB for this G4DDK design when measured with the N4000 noise source.

A comparison of figures 1 and 2 does not show any significant difference between the standard deviations on NF and gain values. We did however get the impression that the standard deviations were significantly lower in the data in figure 2 so we attributed this difference to the low gain of the L LNA and the lower Y-factors that were obtained due to the influence of the NF of the NFA itself. For that reason we measured the L LNA once more. This time with a second L LNA followed by a 3 dB attenuator inserted immediately behind the DUT. The raw data and evaluation is shown in figure 3.

Figure 3. The L LNA measured with the N4000A noise source using a second L LNA directly after the DUT followed by a 3 dB pad to not saturate the NFA.

The temperature on the L LNA was 30.7 C during the measurements in figure 3 so the result at 26 C should be 0.2561 dB. The quarter-wave cable causes a change of 0.0111 dB on the NF when measuring with the N4000 noise source.

The standard deviations on the NF measurements are about 0.002 dB. That means that the standard deviation of an average of 16 values should be 0.0005 dB and the difference between two averages (the influence of the 0.25 wl cable) should have a standard deviation of 0.0007 dB. The true influence of the 0.25 wl cable is best estimated as the average of the results in figures 1 and 3 which is 0.0079 dB. The two measurements of the influence of the quarter-wave cable deviate from this value by 4.5 standard deviations in opposite directions which means that there is a systematic error somewhere. Probably the contact resistance of the SMA connector at the LNA input differs a little from time to time.

The result in figure 3 needs to be corrected by the influence of the second LNA. It is another L LNA with a noise temperature of about 18.5 K. The contribution from it is 0.11 K (0.0016 dB) at the DUT input. With this correction the NF at 26 degrees according to figure 3 should be 0.2545 dB. This compares reasonably well with the result from figure 1 which is 0.2584 dB. The discrepancy is much larger than the statistical uncertainty but since the input connector was not opened and closed between each data point contact resistance variations is a probable cause for the error.

The raw data and evaluation of the SM5BSZ amplifier is presented in figure 4.

Figure 4. The SM5BSZ experimental LNA measured with the N4000A noise source.

The temperature on the SM5BSZ LNA was 27.8 C during the measurements in figure 4 so the result at 26 C should be 0.1363 dB. The quarter-wave cable causes a change of 0.059 dB on the NF when it is measured with the N4000A noise source.

The SM5BSZ experimental amplifier was measured with a L LNA and a 3 dB pad between the DUT and the NFA. The gain is 23.1 dB so the effect of the second stage is 0.09 K or 0.0013 dB. After correcting for the second stage the final result for the NF at 26 C of the SM5BSZ amplifier is 0.1350 dB.

Measurements with the 346A noise head.

The serial number is MY44420333. It was calibrated Feb 27 2006 with this certificate 345a-caldat.jpg

This noise source has an N connector and was used with the same loss before DUT and loss after DUT as the N4000A noise source. Measurements were made the same way as the N4000A measurements, but since the noise source temperature has to be fed to the NFA manually this noise source is less convenient than the N4000A.

Figure 5 shows raw data and evaluation of the AD6IW amplifier followed by another LLNA and a 3 dB pad.

Figure 5. The L LNA measured with the 346A noise source using a second L LNA directly after the DUT followed by a 3 dB pad to not saturate the NFA.

The temperature on the L LNA was 30.8 C during the measurements in figure 5 so the result at 26 C should be 0.2318 dB. The quarter-wave cable causes a change of 0.014 dB on the NF when using the 346a noise source.

The result in figure 5 needs to be corrected by the influence of the second LNA. It is another L LNA with a noise temperature of about 18.5 K. The contribution from it is 0.20 K (0.0029 dB) at the DUT input. (The gain is 2 dB lower with the 346a compared with the N400A.) With this correction the NF at 26 degrees according to figure 5 should be 0.2289 dB.

Figure 6 shows the raw data and evaluation of the SM0ERR amplifier with the 346A noise head.

Figure 6. The SM0ERR LNA measured with the 346A noise source.

The temperature on the SM0ERR amplifier was 28.0 C during the measurements in figure 2 so the result at 26 C should be 0.1830 dB. The quarter-wave cable causes a change of 0.041 dB on the NF for this G4DDK design using the 346A noise source.

Measurements with the 346C noise head.

The serial number is 2339A01657. (calibration info currently missing)

This noise source has a 3.5 mm connector that fits directly to the tested amplifiers. No loss before the DUT when measured directly on the noise source and 0.036 dB loss at room temperature when connected through the quarter-wave cable. The loss after DUT was the same as with the other noise sources.

Figure 7 shows raw data and evaluation of the AD6IW amplifier followed by another LLNA and a 3 dB pad.

Figure 7. The L LNA measured with the 346C noise source using a second L LNA directly after the DUT followed by a 3 dB pad to not saturate the NFA.

The temperature on the L LNA was 30.3 C during the measurements in figure 7 so the result at 26 C should be 0.3050 dB. The quarter-wave cable causes a change of 0.39 dB on the NF when using the 346a noise source.

The result in figure 7 needs to be corrected by the influence of the second LNA. It is another L LNA with a noise temperature of about 18.5 K. The contribution from it is 0.20 K (0.0029 dB) at the DUT input. With this correction the NF at 26 degrees according to figure 5 should be 0.3021 dB.

Figure 8 shows the raw data and evaluation of the SM0ERR amplifier with the 346C noise head.

Figure 6. The SM0ERR LNA measured with the 346A noise source.

The temperature on the SM0ERR amplifier was 28.6 C during the measurements in figure 6 so the result at 26 C should be 0.2380 dB. The quarter-wave cable causes a change of 0.38 dB on the NF for this G4DDK design using the 346C noise source.

Results and conclusions.

Table 2 gives the NF at 26 C case temperature for the three amplifiers with the three noise heads used in this study and with the Orebro setup with circulators which has been corrected for the 0.5 K discrepancy that was detected in Orebro. 

Unit      ----------  NF (dB)  ---------   --- Diff vs Circ. (dB) ---
           Circ.  N4000A   346A    346C    N4000A   346A   345C
SM5BSZ    0.1474  0.1350     -       -     0.0124    -      -
SM0ERR    0.1974  0.1986  0.1830  0.2380  -0.0012  0.0144 -0.0406
AD6IW     0.2674  0.2564  0.2289  0.3021   0.0110  0.0385 -0.0347   
--------------------------------------
Average3  0.2041  0.1967                   0.0074
Average2  0.2324  0.2275  0.2060  0.2700   0.0049  0.0264 -0.0376
Table 2. The NF values obtained above for a case temperature of 26 C and the differences vs the results from the Orebro setup with circulators which is expected to be zero for the N4000A since that was the finding in Orebro.

The Orebro setup was calibrated to show the same result as the N4000A immediately after the Orebro meeting. The errors are very small. The 346A and the 346 C are both close to the N4000A. The noise sources are most probably traceable to the same thermal standard (probably NPL, London) and they may share a common error which however is likely to be very small since it is not very likely that independent methods like sky noise (RW3BP) and ice/water (SM5BSZ) should produce such a close agreement as they do if uncertainties were in the order of 0.2 dB.

The data in table 2 seems to scatter by something like 0.0015 dB. The reason is unknown, it could be variations in contact resistance in SMA connectors but it could also be something else. It is not caused by the statistical vaiations due to the randomness of the noise itself because of the very long averaging times.

The data above suggests that the error when just measuring once with the noise source directly on the LNA is small with the 346A and N4000A noise sources. Worst case in this study is 0.03 dB for the neutralized SM5BSZ amplifier which has near infinite VSWR. It is possible that the influence of the quarter-wave cable would have been much different with an adapter that would have changed the phase angle of the impedance change in the noise source however. Nevertheless it seems reasonable to give NF values with two decimals when using modern standard equipment. Particularly when comparing similar amplifiers.

It is NOT acceptable to tune amplifiers on the standard instruments however. Tuning has to be done some other way. One can use FM quieting on a weak carrier. Fast and easy if the system temperature is low as it is when the weak signal comes from an antenna pointed into the sky. Another possibility is to use a SINAD meter on a carrier in SSB mode or better on a FM modulated signal. One can also measure S/N for example with Linrad in which case tuning is reasonably fast even with a room temperature source.