This page is a sub-page to
PRECISION MEASUREMENTS OF NOISE FIGURES.
It gives raw data and a detailed description of the procedures
used to measure the NF (and gain) of a set of low noise 144 MHz
amplifiers on December 28 2012.
The measurements were made by Mart SM0ERR and Leif SM5BSZ.
The test objects.The amplifiers are listed here Tested amplifiers in November 2012. with links to network analyzer plots of matching and transmission both forwards and backwards.
All the amplifiers are designed with BNC connectors. Unknown differences in loss between different adaptors does therefore not come into play.
In all six complete series of measurements were made on six of the amplifiers. Two of them, the AD6IWW and the ATF33143november are not stable with a circulator directly on the input so they were not included in the evaluation of the offset between the NF scales even though they were measured in the two configurations where the circulator was not used. The NF and gain of those is accurately determined anyway.
Equipment.Noise figure meter: Agilent 8973A.
Noise head: N4000A. S/N MY4420135 (Calib 19-apr-2012.)
Circulator: A dual circulator with a forward loss of about 0.45 dB and better than 75 dB backwards isolation. The return loss was about 26 dB towards the DUT. Z=(54.096,-0.701)
Impedance tuner: A 2 m long 0.5 inch flexwell cable which was squeezed for the impedance seen by the DUT to be 50 ohms on 144 MHz with the cable connected to the circulator.
Freq Impedance SWR RL (MHz) (Re,Im)Ohm (Ratio) (dB) 142 (50.58, 1.52) 1.033 35.7 144 (49.96, 0.02) 1.001 ~70 146 (48.80,-0.92) 1.031 36.1
Adaptor, N female to BNC male: Good quality with gold pins.
Adaptor, N female to N female: Good quality with gold pins.
Cables:82 cm RG223 with N male in one end and BNC male in the other end. Nominal loss 0.129 dB.
Quarterwave: 33.5 cm RG223 with BNC male and BNC female. Nominal loss 0.526 dB.
Calibration.For all measurements the instrument was calibrated with the noise head directly on the NF meter. The attenuation of the cable used between the DUT and the NF meter was fed into the NF meter together with the ambient temperature.
For measurements 1,2 and 7 the loss of the combination of the circulator and the 82 cm cable, 0.600 dB was fed into the NF meter. The temperature of the circulator was monitored and the data was changed from time to time. The temperature varied less than 1 degree during the whole session.
For measurements 5 and 6 the loss of the combination of the circulator, the impedance tuner, the N to N adaptor and the 82 cm cable, 0.657 dB was fed into the NF meter.
Measurement procedure.The 8973A was set to make 256 averages at 5 frequencies with 10 kHz separation. The table of 5 independent gain and NF values were fed into Excel where they were averaged and where also the standard deviation was computed. This way the total averaging was 1280 and we also got a check for mistyping errors.
Evaluation.The evaluation spread sheet can be inspected in HTML format here and downloaded in Excel 97-2003 format as nf2012dec.xls as well as in Open Document format as nf2012dec.ods
The fields with a green background colour contain the raw data used to evaluate the absolute NF scale by a comparison with the data obtained by use of ice and steam in previous experiments. The ice and steam values are in fields with a red background colour.
The raw data is in groups of five data points and the standard deviation for an individual measurement is 0.0025 dB (RMS for all 36 standard deviations.) This means that the standard deviation for an average of five measurements should be 0.0011 dB.
The Z dependence, the difference between the result with and without the quarterwave cable should be zero for configurations 5 and 6 because in this case the quarterwave cable causes an extremely small change of the impedance. By selecting a loss for the quarterwave cable that makes the average dependance of whether the cable is inserted or not zero we arrive at a loss for inserting the cable of 0.0601 dB. That is higher than the 0.0561 to 0.0563 dB that we obtained from the network analyzer and significantly more than the nominal 0.526 dB for the cable. Attributing 0.075 dB to the two connectors does not seem right, but the standard deviation of the Z dependance is only 0.0018 dB which is in close agreement with the expected standard deviation between two averages of five individual measurements. The study of very small losses is a separate subject of great interest for precision NF measurement...
Accepting 0.601 for the loss in the quarterwave allows the determination of NF values as the average of the measurement with and without the quarterwave cable. These averages have a blue background colour in the spread sheet. The standard deviation for such averages should be 0.0078 dB. These NF values come in 3 groups and they are accurate NF results on a relative NF scale.
By comparing the three tables of relative NF values with the data previously obtained with Linrad and the hot/cold method using ice and steam it is possible to find by how much the NF scales have to be shifted to make the average NF for all the amplifiers equal. At this point we have an unexpected result. The shifts needed are very small
To make measurements 3 and 4 fit 5 and 6 it is necessary to change the loss of the stuff on the input side of the DUT from 0.0657 to 0.0652 dB. That is FINAL DIFF VS HOTCOLD which is 0.0253 and 0.0303 respectively. Assuming the loss really was 0.652 for the stuff between the noise head and the DUT we find that the calibration error is 0.0253 dB only. Now, that is the sum of the calibration error for the N4000A noise head specified as 0.16 dB and the ice/water experiments by SM5BSZ which are believed to have error limits of 0.04 dB. It should be noted however that the Agilent specification is for a single measurement, not for the average of measurements with and without a quarterwave cable.
The standard deviation for the differences between the NF values obtained in the three series of measurements are 0.0047 dB for 1,2 vs 3,4, 0.0064 dB for 1,2 vs 5,6 and 0.0025 dB for 3,4 vs 5,6. This means that configurations 3,4 and 5,6 give very similar relative NF values. After accepting a change from 0.0657 to 0.0652 dB for the stuff between the noise head and the DUT they are almost identical. The standard deviation for the differences between relative NF values should be 0.0015 dB (the square root of two times the standard deviation of each NF value 0.0011 dB.) The measurement series 1,2 is more like the ice and steam measurements than 3,4 and 5,6 with a standard deviation of 0.0023 dB only.
According to the series 2,3 which does not have anything more than a N to BNC adaptor between the noise head and the DUT the ENR value specified by Agilent is 0.0253 dB plus the unknown loss in the BNC to N adaptor too high. The series 1,2 suggests the same difference to be 0.0275 dB but it contains the uncertainty in the loss of the circulator and cable and also the unknown loss of the N to BNC adaptor.
Adopting the shifts AVG AVG DIFF VS HOTCOLD and shifting the relative NF values (blue background) by that amount gives the final absolute NF values of the three measurement series. That is the values to the immediate right of the blue fields. The average of those absolute NF values are listed for each amplifier on the NF line with a yellow background colour and represent the final results of this study.