NOISE SOURCES COMPARISON AROUND THE EUROPE

Transcription

1 INTRODUCTION It s well known that the orret way to furnish an experimental measure is to provide both the observed value and its assoiated unertainty. For the low noise amplifiers, the improvement of the tehnology and the ryogeni ooling has produed a dereasing of the input Noise Temperature down to few times the theoretial quantum limit hν/2, while the assoiated unertainty isn t affeted by any redution. It isn t rare a ase we quote the noise temperature of an LNA as Te = 10 +/-20 K. The auses that generates and propagates the Unertainty are well desribed and disussed in [1] and [2]. Here s a summary of that auses and typial values for the travelling LNA measured by a typial mod. 346C Noise Soure [3] [ db ] Unertainty related to the LNA Noise Figure [strongly dependent on U(ENR)] 0.20 Unertainty related to Instrument Noise Figure Unertainty related to LNA Gain Measurement 0.01 Unertainty related to ENR 0.15 Overall (RSS) Unertainty on Noise Figure (1 σ) 0.25 In this ase, the NF of the LNA is NF = 1.7 db +/-0.25 db or 139 +/-26 K ( 1σ). It s evident that the strongest ause of U(Te) is the U(ENR). The most obvious questions is how the U(ENR) an be redued? Before to answer, it s neessary give the answer to another question: What is the minimum theoretial limit of U(ENR)? The answer may be found in [4] and in [5], and may be summarised as follow: ENR is determined by the help of a radiometri hain. At the input a relays swith sequentially three arms of the hain, the old standard, the hot standard and the unknown noise soure. U(ENR) is the resultant propagation of many auses of unertainty but the most signifiant are: The unertainty of the temperature of the old noise soure U(Told) The unertainty of the spar of the arms of radiometri hain used for omparison U(sij). 1

2 Given the most aurate values obtainable from ommerial Network Analyzer operated by a very skilled operators, a Noise Generator may be alibrated with U(ENR) = +/ db (1 σ) at 18 GHz. For more than 20 years, the hp/agilent most aurate Noise Generators, mod. HP346 had U(ENR) quoted +/-0.15 db (1 σ). In the hp/agilent douments, for the mod. HP346 has been wrote that: the vetor method to redue the U(ENR) was known but not applied [6] the unertainty was quoted at 1 σ instead the more ommon adopted value 2 σ [7]. In Year 3, a join workgroup of Institute of Radioastronomy and University of Tor Vergata has lowered the unertainty of their own Noise Generators by appliation of the most aurate vetor method known [5]. The unertainty reahed was U(ENR) = +/ db (1 σ) at 18 GHz. 1 In house Noise Soure alibration with seondary standard In year 4 Agilent started to sell a more aurate Noise Generator, the model series SNS N400x, the related Unertainty was quoted: U(ENR) = +/-0.06dB (1σ). Despite of the unertainty has been strongly redued and the theoretial limit has been almost reahed, the value of db annot be onsidered a small value. Note that realistially, a U(ENR)= db one propagated, produes a U(NF) = 0.09 db ( +/-7 K at 1 σ or +/-14K at 2 σ) [3]. What ation an be taken in order to further improve the unertainty? As suggested by [8] and [9] the only possible ation is a omparison. The inter-laboratory omparison by itself doesn t further redues the unertainty but it enhanes the differenes and should suggest some orretive ations, espeially for the values farther from the mean. 2

3 COMPARISON The inter-laboratory omparison is not a re-alibration. By itself doesn t hange the ENR of the ompared Noise Soures. The differenes in readings ourring at eah laboratory will not be leared. Only one this doument has been read, eah laboratory will have the freedom to apply orretions to ENR. In any ase, eah orretion should be applied with great preaution and autiousness. The anonial way to do a omparison is a omparison of a travelling standard. An Input Noise Temperature Standard ( an LNA with a standard Noise Temperature) doesn t exist. Te of LNA is every time dependent on ambient temperature, bias supply and even on the refletion oeffiient of the soure. But if the travelling LNA will be measured at known ambient temperature, biased by a very stable power supply and lose on well mathed load, its Te an be onsiderate stable enough around the travel. The omparison is more aurate as smaller is Te of the LNA. So a ryogeni omparison, in priniple, should be preferred to a room temperature omparison [10]. But pratially speaking, a balaning of the Pros & Cons must be done: Cryo omparison LNA Room temperature LNA omparison Man power required week (s) ( - ) hours ( + ) Auray best ( + ) high ( ) Sensitivity to how the setup is arranged more ( - ) less ( + ) Sensitivity to U(ENR) less ( + ) more ( - ) Easy to do less ( - ) more ( + ) Probability to find andidate laboratories less ( - ) more ( + ) Due to that reasons, a room temperature omparison has been preferred. Frequenies: Often the frequeny boundaries omes at 18 GHz, 26.5 GHz, 40 GHz, 50 GHz, 67 GHz The best seller Noise Generators overs the very popular bandwidth 10 MHz-18 GHz. Of ourse, higher frequenies are also very important for radio astronomy. It has been hosen a frequeny range GHz for the following reasons: The high end of the bandwidth (say 18 or 26.5GHz) is probably affeted by greater unertainty or errors than the low end of the bandwidth. A omparison is probably more meaningful at the high end of the bandwidth GHz has been hosen beause step over the 18 GHz boundary. In this way, even the laboratories that are arranged up to 18 GHz may takes advantage of the omparison. 3

4 Refletion oeffiient and impedane mathing: The LNA has been impedane mathed as well as possible, but it annot be regarded as a well mathed devie. As higher is the produt of the refletion oeffiient ( ρns * ρlna) as higher is the ripple of the Te urve over frequeny. It is request to the laboratories to perform two measurements, one plaing a seleted low VSWR, ferrite isolator in front of the LNA and the other one without that irulator. All the measurements made with the irulator will be ompared all together, as well as the measurements performed without the irulator. Comparing the two obtained measurement traes, it s possible distinguish if the ripple is mainly due to mismath errors or due to ENR differenes. Drift: Sine a drift on gain and noise temperature may our during the measurement session, a measurement of the thru is requested both at the beginning and at the ending of the measurement session. If an important drift will be seen, the Te data file will be orreted. Room temperature: The handling of the Noise Soure and LNA as well as the solar light oming trough a window, as well as the viinity of an heater or air onditioning hanges the physial temperature of the LNA and onsequently it s Te. For this reason a proper shield form heat/old soures is required, as well a period of time long enough to reah the thermal equilibrium ( 5 minutes). Aging hek: At INAF-IRA the LNA has been measured both at the beginning and one year later at the end of the ampaign. Differenes on Te was smaller than +/- 2 K exept for frequenies greater than 26.3 GHz where the differene was slowly inreasing with the frequeny up to 10K. 4

5 DATA PROCESSING The data files oming from eah one laboratory have been analyzed. Not all data set are organized on the requested way. Some data are 50 MHz spaed while other are MHz spaed. One data set is 500 MHz spaed. This differene has involved a small ompliation on the proessing but does not affets the omparison. Most data set ontains Noise Temperature and Insertion Gain while two data set ontains NF (db) only. This small lak does not affet the omparison. Some data set do not inlude the initial and final zero traes. It has been assumed that the drift was negligible. Where possible, the drift has been evaluated. For all data set both the Noise and Gain drifts were negligible. Sine the data set have been olleted at different physial temperatures, a orretion has been applied in order to normalize the atual Noise Temperature. The experimental found law ΔTe =+ 0.5 K / C has been applied. ΔTamb Two families of Cartesian graphi have been plotted: the ase without input isolator (NO_CIRCULATOR) and the ase with the isolator in front of the LNA (YES_CIRCULATOR). As it may be seen, the spread is very large. Also it s diffiult to understand the whole traes. A seletive analysis is muh more indiative. Three kind of analysis has been performed: The effet of the mismath loss is analyzed and disussed. the atual Noise Temperature trae is ompared to the mean value (alulated at every frequeny along the points oming from eah Noise Generator). Traes performing similar behaviour has been looked for, grouped, analyzed and disussed. What kind of mean? The most known method to estimate the mean value of a number of samples is the alulation of the arithmeti average, aomplished by +/- the standard deviation. An alternative method, widely used at NIST, is the omputation of the median aomplished by +/- the MAD [11]. For uniformly and normal distributed samples, the average and median are oinident, as well as the standard deviation and the MAD. But in the ase that few samples fall far outside the distribution law, the median better filters away the farther samples. Even the MAD, rather than the standard deviation, better filters the rare and farther samples. So the median method is more robust than the average one. 5

6 RESULTS Impedane Mismath effets A mismath error should generates a ripple on the frequeny domain representation of the trae. The mismath error is as large as the trae NO_CIRCULATOR shows more ripple than the trae YES_CIRCULATOR. The mismath effet has been summarized on the next table: ORGANISATION Mismath error NOISE Freq. (differene on GENERATOR range the ripple) INAF-IRA HP 346C +10 db absent INAF-IRA HP 346 n/a Uni Cantabria HP 346C K01 strong Uni Cantabria HP 346C K01+10 db absent Uni Manhester SNS N4002 small Yebes HP 346C+10dB absent Yebes NC 346KA+10dB absent Yebes SNS N4002 small Astron HP 346ª n/a Astron SNS N4000 n/a Fraunhofer IAF SNS N4000 n/a Fraunhofer IAF SNS N4002 absent

7 Charts Noise Temp. of the same LNA measured by different labs. (NO_input_irulator) Noise Temperature (Te) [ K ] IRA 346C+10dB IRA 346A U.Cantabria 346CK01 U.Cantabria 346CK01+10dB U. Manhester N4002A Yebes 346C+10dB Yebes 346KA+10dB Yebes SNS N4002 ASTRON 346A ASTRON SNS N4000A Fraunhofer IAF SNS N4002A Fraunhofer IAF SNS N4000A Frequeny [ GHz ] 2 Noise Temperature of same LNA measured by different labs ( NO_input_irulator) Noise Temp. of the same LNA measured by different labs. (YES_input_irulator) Noise Temperature (Te) [ K ] IRA 346C+10dB IRA 346A U.Cantabria 346CK01 U.Cantabria 346CK01+10dB U. Manhester N4002A Yebes 346C+10dB Yebes 346KA+10dB Yebes SNS N4002 ASTRON 346A ASTRON SNS N4000A Fraunhofer IAF SNS N4002A Fraunhofer IAF SNS N4000A Frequeny [ GHz ] 3 Noise Temperature of same LNA measured by different labs ( YES_input_irulator) 7

8 Noise Temp. of the same LNA measured by different labs. (YES_input_irulator, GHz only) Noise Temperature (Te) [ K ] IRA 346C+10dB IRA 346A U.Cantabria 346CK01 U.Cantabria 346CK01+10dB U. Manhester N4002A Yebes 346C+10dB Yebes 346KA+10dB Yebes SNS N4002 ASTRON 346A ASTRON SNS N4000A Fraunhofer IAF SNS N4002A Fraunhofer IAF SNS N4000A Frequeny [ GHz ] 4 Noise Temperature of same LNA measured by different labs ( YES_input_irulator, GHz only) Even if the different olours may help the trae identifiation, the whole traes are not the best way to ompare different measures beause is onfusing. A good way to ompare eah measurement is a plot of every measurement ompared with the mean of the measurements. As disussed before, the median value, rather the average, better desribe the most probable mean. 8

9 NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 5 INAF-IRA, HP 346C+10 db attenuator NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 6 INAF-IRA, HP346A 9

10 NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 7 University of Cantabria, HP346CK01 NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 8 University of Cantabria, HP346CK01+10dB Attenuator 10

11 NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 9 University of Manhester, SNS N4002A NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD NO_ir_ 10 Centro Astronomio de Yebes, HP346C+10dB Attenuator 11

12 NO_ir_avg_or Te YES_ir_avg Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 11 Centro Astronomio de Yebes, NC346KA+10dB Attenuator NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 12 Centro Astronomio de Yebes, SNS N4002A 12

13 NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 13 Astron, HP346A NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 14 Astron, SNS N4000A 13

14 NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 15 Fraunhofer IAF, SNS N4002A NO_ir_avg_or Te YES_ir_avg_or Te MEDIAN+/-MAD NO_ir_ MEDIAN+/-MAD YES_ir 16 Fraunhofer IAF, SNS N

15 Trae behaviour, similarities and noise generators summary In the following table the behaviour of the ompared noise generators and the unertainty labelled by the fatory are summarized ORGANISATION NOISE Date Freq. Similar Unertainty GENERATOR range behaviour u(enr) INAF-IRA HP 346C +10 db 03/09/ CM +/ (1) INAF-IRA HP /08/ / Uni Cantabria HP 346C K01 20/07/ / (2) Uni Cantabria HP 346C K01+10dB 20/07/ CM +/ (2) Uni Manhester SNS N /10/ / Yebes HP 346C+10dB 30/11/ CM +/ Yebes NC 346KA+10dB 30/11/ / Yebes SNS N /11/ / Astron HP 346A 20/04/ CM +/ Astron SNS N /04/ / Fraunhofer IAF SNS N /05/ CM +/ Fraunhofer IAF SNS N /05/ / INAF-IRA HP 346C +10 db 08/06/ Repeatibility hek INAF-IRA HP /06/ Repeatibility hek Where: (1) Unertainty redued by in-house alibration (2) K01 stands for 1GHz 50GHz 2.4mm oax onnetor. the behaviour of the traes is similar among themselves the behaviour of the traes is similar among themselves CM the trae is loser to median value It has been looked for the most similar urves. In the following two graphis are plotted le losest urves beause the number of similar families of urves is two. 15

16 As it may be seen in the following harts, the measurements performed by laboratories that involves the mod. NC346C + 10 db attenuator are loser among themselves. Also measurements performed by laboratories that involves the SNS N4002 are among themselves. But the two families are different among themselves. It may be suspeted that the differenes may be imputed to differenes in the original alibration proedures at Agilent fatory IRA_346C+10dB Uni_Cantabria 346CK01+10dB Yebes_346C+10dB AVG seleted over 346C's Measurements performed with Noise Generators mod. 346C+10 db Attenuator by different labs U_Manhester_SNS N4002A Yebes_SNS_N4002A Fraunhofer_IAF_SNS_N4002 AVG seleted over SNS_N4002A's Measurements performed with Noise Generators SNS_N4002A by different labs 16

17 ENR TABLE CORRECTION For laboratories like to orret their ENR table in order to redue the distane from the average or median, a method is desribed. It s important to write in an expliit way that this operation is oneptually a mistake for two reasons: A metrologial mistake. The alibration of a Noise Soure by the knowledge of Te of LNA is same as the alibration of a poket rule by omparing to the length of a table desk (while the right way is the ontraire). A mathematial mistake. It isn t a rigorous method, it s just s an approximation. The reason is beause for a rigorous alulation it should be neessary to know and take into aount for the noise temperature of the Noise Figure Meter ( the Calibration data), while those data are unknown. However sine the LNA gain is high enough, the Noise Temperature of the Noise Figure Meter is strongly masked, so aepting a small error, the Noise ontribution of the noise figure analyzer may be negleted. In any ase, even if it s an approximation the method is onverging, so the error will be redued and there is no risk of overorretion. But if the differene of ENR is suspeted to be very large, a orretive ation is perhaps better than do nothing. So, if the preeding simplifiation is aepted, the atual measured Noise Temperature is: Th Y T Te = (1) Y 1 The similar equation form may be written for the value we would read, the orreted Noise Temperature ( Te) Th YT Te= (2) Y 1 Now, let perform the ratio, Te Th YT = Te T h Y T (3) Solving for the orreted hot temperature Th, Th Th Y T = + YT (4) Te Te 17

18 Finally the new, orreted ENR value is: Th To ENR = 10 log( ) (5) To Or, if the operator prefers the advantages of the small orretion for T To, even if formally different by the ENR definition, we get: Th T ENR = 10 log( ) (5 bis) 290 Where: Te = Input Noise Temperature of the LNA ( the atual Noise Temperature) Th = Hot Noise Temperature generated by the Noise Generator (the atual value) To = 290 K, The standard temperature Y = The Y fator. Te = Input Noise Temperature of the LNA ( the wanted value) Th = Hot Noise Temperature generated by the Noise Generator (the wanted value) ENR = The Exess Noise Ratio (the atual value) ENR = The Exess Noise Ratio (the wanted value) Note that the variables Te and has a physial meaning. Te should not only be onsidered alone, but also their ratio The knowledge of the variable Y is required but probably was not reorded at the time of measurement. It doesn t matter beause the Y fator may be alulated offline by solving the (1) for Y and ombining the (5) or the (5bis) Te Te ENR 10 Th = To + To 10 (6) Y Te + To + To 10 = Te + T ENR 10 (7) In order to help users, who want orret their ENR table, a MS Exel file is provided. The user should input, frequeny by frequeny, the atual ENR, the atual Te and the spreadsheet will alulate the ENR orreted that produe the median Te. 18

19 CONCLUSIONS: The Noise Temperature measurement of a travelling LNA has shown strong differenes among laboratories. Assuming as the major ause of unertainty on Noise Temperature the unertainty of ENR, the spread of the data is muh larger than the expeted. Many of the Noise Soures were lowest unertainty type, so there is no room for improvements involving the fatory. The only way to further redue the unertainty of the ENR is the alibration with a ryo-load that ats as a primary or seondary metrologial standard. This is espeially thru for the millimetre wave frequenies where the unertainty is greater. Beause the ryo loads are very expensive ( ) a join or onsortium is suggested for the purhase, as well as a sharing in the use. Perhaps the best way would be a ommon European Centre for Noise Soure Calibration that involve metrologial standard old-loads as shown in the following pitures. Noise/Com Cold Load Maury Mirowave Cold Load In any ase, eah laboratories that believe its ENR is wrong, has the freedom to orret it by using the algorithm provided. ACKNOWLEDGEMENTS Thanks to all very skilled and motivated partiipants, espeially to Eduardo Artal and Juan Luis Cano ( Uni of Cantabria, Spain ), Juan Daniel Gallego ( CAY, Spain ), Mohamed Missous and Shahzad Arshad ( Uni of Manhester, U.K. ), Eri Van Der Waal ( Astron, NL ), Hermann Massler and Beatriz Aja ( Fraunhofer IAF, Germany ), Federio Perini ( INAF-IRA, Italy ). 19

20 BIBLIOGRAPHY [1] N. Kuhn; Curing a Subtle but Signifiant Cause of Noise Figure Error, Mirowave Journal,; June 1984 [2] AN-57-2, Noise Figure Measurement Auray The Y Fator, hp/agilent Tehnologies year 1 [3] [4] W.C. Daywitt; ;Radiometer equation and analysis of systemati errors for the NIST automated radiometers, NIST Tehnial Note 1327, marh 1989 [5] M. De Dominiis; Strumentazione e Metodologie per la Modellistia di Rumore di Dispositivi Attivi ad Alta Frequenza, PhD Thesis Uni. Tor Vergata, February 4 ( in Italian) [6] AN-57-1, Fundamentals of RF and Mirowave Noise Figure Measurements, hp/agilent Tehnologies, July 1983 [7] Noise Soures 10MHz to 26.5GHz, Hewlett Pakard Tehnial Data, Jan [8] [9] J. Randa, and others; International Comparison of Thermal Noise-Temperature Measurements at 2, 4, and 12 GHz e, IEEE T.I.M. APRIL 1999 [10] J.D. Gallego, J.L. Cano, Estimation of Unertainty in Noise Measurements Using Monte Carlo Analysis:, Radionet FP7 Workshop Meeting, June 9 [11] 20