The three amplifiers brought to the meeting by RW3BP were
measured with the Agilent 8973 noise figure meter and a N4000A
noise head that was calibrated by Agilent on May 20 2013.
The
Calibration Report
gives the impedances as (49.57, j0.26) ohms for SOURCE ON and
(49.35, j0.18) ohms for SOURCE OFF.
The magnitude of the difference is 0.24 ohms which is a 0.48%
impedance change.
As can be seen below even a modest change like that has a significant
influence on the result.
First the amplifiers were measured with and without a quarter-wave cable. See figure 1. Here the amplifiers were measured cold. Just powered up and measured immediately. |
Figure 1. Agilent 8973A with the N4000A noise head measuring RW3BP amplifiers with and without a 0.25 wl cable. For the RW3BP-6 amplifier the NF change is 0.050 dB for the measurement with and without the quarterwave cable (the loss of the cable is about 0.0345 dB). Thus the error caused by the 0.5% impedance change of the noise head is 0.025dB for this amplifier. There are amplifiers that are far more sensitive to the feed impedance. |
After that the amplifiers were measured with
two circulators and a 10 dB pad which was selected
to present 50.0 ohms to the test object.
With so much attenuation the Y-factor becomes low
and for that reason the NF was measured 10 times to produce
a good average.
The standard deviation in individual measurements is
about 0.006 dB so the standard deviation of the averages
should be about 0.002 dB.
See figure 2.
|
Figure 2. Agilent 8973A with the N4000A noise head measuring RW3BP amplifiers through circulators and a 10 dB attenuator. |
Table 1 shows the results as NF and Te.
|
8973A 8973A 8970A RW3BP 0.25 wl Circulator Circulator Sky temp Unit Te NF NF NF-10.64 dB Teraw26 Te NF (K) (dB) (dB) (dB) (K) (K) (K) (dB) RW3BP-13 9.60 0.1414 10.7843 0.1443 9.80 8.74 9.85 0.145 RW3BP-12 10.60 0.1559 10.8170 0.1770 12.06 8.51 10.19 0.15 RW3BP-6 8.74 0.1290 10.7914 0.1514 10.29 8.83 10.19 0.15 |
Table 1.The measured data on RW3BP amplifiers. |
The noise temperature of a LNA depends on the physical temperature
of the unit.
The slope for a DDK amplifier is 0.13 K/degree C.
To compare the different series of measurements it is necessary to
compensate for the different temperatures.
The 8973A measurements were made with cold amplifiers at 26 C.
The RW3BP measurements were made at an ambient temperature around 25 C
but without a fan cooling the amplifier.
This means that the RW3BP data probably is valid for a box temperature
of 32 C.
The Teraw26 data correspond to a box temperature of 27 C.
The RW3BP-12 amplifier was left running while the circulator and attenuator was connected. There were some things to discuss so it may have been running for ten minutes. The amplifier was lying on the table with poor air circulation and it might have rised by 1 K in Te (7.7 degrees in box temperature.) |
8973A 8973A 8970A RW3BP Unit 0.25 wl Circulator Circulator Sky temp (K) (K) (K) (K) RW3BP-13 9.60 9.80 8.67 9.07 RW3BP-12 10.60 11.06 8.38 9.41 RW3BP-6 8.74 10.29 8.70 9.41 |
Table 2.The data in table 1 corrected for temperature differences as described in the text. |
The circulator measurements only give the noise temperature
on a relative scale.
They do however not agree on the RW3BP-12 amplifier.
That could be due to an error in the guess of the temperature
of the box in the 8973A measurement.
There could also be a connector problem.
We did not clean connectors
because of the very long drying time required.
The measurements with and without a 0.25 wl cable carry the absolute calibration of the noise head albeit with added uncertainties due to the uncertainty in the loss of the quarter wave cable and the fact that the quarterwave cable is not exact in length and impedance. The loss of 0.0345 dB in the quarterwave cable comes from insertion loss measurements. The same cable was subsequently measured by SM0ERR using the S11 method with the result 0.0355 dB. A big study of NF vs source impedance gives the loss of the quarterwave cable as 0.0338 dB. It seems likely that this value is more correct than the old value 0.345 used in figure 1. A 0.0007 dB lower loss in the quarter wave cable means that 0.02 K should be added to Te. The 8973A measurementd use a Nf-SMAm adapter which is likely to add 0.63 K [3] This means that 0.63 K should be subtracted from Te. After applying these corrections table 3 is obtained. |
8973A 8973A 8970A RW3BP Unit 0.25 wl Circulator Circulator Sky temp (K) (K) (K) (K) RW3BP-13 8.99 9.80 8.67 9.07 RW3BP-12 9.99 11.06 8.38 9.41 RW3BP-6 8.13 10.29 8.70 9.41 |
Table 3.The data in table 2 corrected for adapter and quarterwave cable losses. |
The effect on the NF reading when inserting the quarterwave
cable is 0.050 dB for RW3BP-6 while it is 0.028 dB for
RW3BP-12 and 0.024 dB for RW3BP-13 (see figure 1.)
That is not unexpected it depends on the return loss and it
seems reasonable to compare the N4000A calibration to the RW3BP
sky noise calibration by use of the result for RW3BP-13 only.
The conclusion is that the sky noise observations by RW3BP
are in full agreement with the N4000A calibration.
The near perfect agreement in table 3 is presumably accidental,
but it very strongly suggests that something is wrong with the
error estimates in the Agilent
Calibration Report
That is a statement made with the findings here
Precision Measurements of Noise Figures
in mind.
Measurements with ice and boiling water on 144 MHz showed an
agreement that was nearly one order of magnitude better than expected.
It seems to me that the error evaluation in the calibration
procedure is not taking into account that some errors cancel.
If the measurement error due to impedance variations is 0.025 dB as we find for the RW3BP with the N4000A noise head a conservative error estimate for the comparison of two different N4000A noise heads would be 0.05 dB. One might however argue that the error of the difference would be 0.035 dB with the argument that the errors are uncorrelated. In real life however, the impedance changes are likely to be similar and the errors would largely cancel. I think that is the explanation why the results seem far more accurate than expected. It seems to me that it should be possible to re-evaluate the error limits because the impedance changes of the different standards back to metrology institutes should be available. The noise temperatures obtained from the 8970A with circulators seem to be too low by 0.5 K. |