PROCEEDINGS OF SPIE - ntut.edu.twwangsen/Courses/A CMOS... · design and implementation of mm-wave radiometers with excellent SiGe demonstrations at frequencies up to 100 GHz with

PROCEEDINGS OF SPIE

SPIEDigitalLibrary.org/conference-proceedings-of-spie

Overview of CMOS technology forradiometry and passive imaging

Adrian Tang

Adrian Tang, "Overview of CMOS technology for radiometry and passiveimaging," Proc. SPIE 10194, Micro- and Nanotechnology Sensors, Systems,and Applications IX, 101942P (18 May 2017); doi: 10.1117/12.2260920

Event: SPIE Defense + Security, 2017, Anaheim, California, United States

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 10 Mar 2019 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use

Overview of CMOS Technology for Radiometry and Passive Imaging

Adrian Tang Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, 91109

ABSTRACT

This paper discusses the effects of gain drift and variation often referred to as “ΔG/G” in CMOS mm-wave radiometers as well as the techniques employed to suppress its effects on radiometric and passive imaging instruments. The paper presents a demonstration of a CMOS Dicke-switched radiometer which uses correlated double sampling to eliminate gain variation, and investigates the contributions of the ΔG/G behavior to the demonstrated instrument’s measured NEΔT through experimental approaches. Finally the paper describes how the discrepancies between calculation and measurement can be attributed to limited performance of Dicke switches in silicon and how further development is needed to truly hit useful NEΔT resolutions in the 1ºK range. Keywords: millimeter-wave imaging, passive imaging, Radiometry

1. INTRODUCTION Although radiometry is a relatively new topic for the commercial silicon (CMOS & SiGe) community, radiometric observations however have a 100+ year-long heritage in the scientific community, most notibly in areas of radio-astronomy, weather observations, and more recently planetary science. Of all the available wavelengths, radiometric observations performed at millimeter-wavelengths remain an invaluable tools for measuring Earth’s climate by providing a means to measure the distribution of water vapor inside severe storms and hurricanes, and for tracking pollutant distributions over the planet as they are carried by wind or storm systems [1,2,3]. For planetary exploration, mm-wave radiometry provides a means to detect liquid water on comets or planetary bodies, as well as study atmospheric constituents, measure wind velocities and even surface properties like grain size or roughness [4].

For the silicon community however, the growing interest in mm-wave radiometry is focused on potential commercial applications including security screening of persons at stand-off distances, using passive imaging to see through fog or dust for self-driving cars, and product manufacturing / quality control of industrial materials. Indeed the emerging silicon mm-wave community has made great progress in the design and implementation of mm-wave radiometers with excellent SiGe demonstrations at frequencies up to 100 GHz with compelling noise performance [5,6], and CMOS demonstrations at similar frequencies with more modest noise performance, but offering the added benefit of higher levels of integration with ADCs and other back-end processing elements [7,8,9,10]. While these radiometer receivers have been demonstrated in silicon with impressive performance, few have been employed or evaluated to perform actual radiometric observations, and so several subtle effects unique to the high sensitivity and long-term stability requirements of real-world radiometric observations are yet to be considered by the silicon community. The purpose of this paper is to explore these considerations in the hope of encouraging even more progress. Of these radiometric effects the most notable is gain variation, often called ΔG/G for short, which can drastically affect sensitivity if not well mitigated and calibrated within a radiometer, which will be explored in this paper.

Invited Paper, Rising Researcher Paper

Micro- and Nanotechnology Sensors, Systems, and Applications IX, edited by Thomas George, Achyut K. Dutta, M. Saif Islam, Proc. of SPIE Vol. 10194, 101942P · © 2017 SPIE

CCC code: 0277-786X/17/$18 · doi: 10.1117/12.2260920

Proc. of SPIE Vol. 10194 101942P-1Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 10 Mar 2019Terms of Use: https://www.spiedigitallibrary.org/terms-of-use

2. ACHIEVING HIGH RADIOMETRIC SENSITIVITY For radiometric observations the key desired characteristic is “high sensitivity” which can be quantified

through several different metrics, some of which do and do not depend on how a radiometer is operated. Noise Equivalent Power or (NEP) is a relatively “operation independent” assessment of sensitivity and describes the input referred noise power of a receiver or detector when the detector is “observed” or averaged for exactly 1 second. The NEP value is often applied to detectors (pre-amplified or stand-alone) and provides a solid basis for performance comparisons as NEP quantities capture both the RF bandwidth and the responsivity or “power-to-voltage” coefficient, (responsivity is similar to gain of an amplifier except that a detector is power in and volts out so there are unit differences) of the detector. However for radiometry, NEP does not directly translate into a quantity that is intuitive, and so noise equivalent delta temperature (NEΔT) is often used to describe a radiometer instrument’s overall performance. NEΔT describes the smallest temperature difference between two targets that a radiometer can resolve with a given integration time, or equivalently: “the statistical variance in estimated target temperature when measuring the same target repeatedly over the same observation period” (since differences smaller than the measurement variance cannot be resolved). The NEΔT can be readily expressed from receiver and other RF system parameters with the classic radiometer equation [11]:

∆ = 1 + (∆ )

Which describes the relationship between NEΔT, the noise temperature of the receiver system, Tn, the bandwidth being observed B (which has different meanings depending if the radiometer is a heterodyne or direct detection based), τ the observation time of the radiometric measurement, and an additive term ΔG/G which describes gain variations in the receiver (described in the next section). Looking into the radiometer equation, if one desires to improve the NEΔT of an instrument, several design choices exist. A longer observation or integration time will reduce the first term and lower the NEΔT, but only to the point where the second ΔG/G term becomes dominant. This relationship is very intuitive as statistically averaging a power signal will increase the robustness of the power estimate by √T. However ΔG/G is a non-zero-mean noise and so even further averaging may actually inflate this term, overcoming the benefits (discussed in the next section). Similarly increasing bandwidth provides a lower NEΔT as a wider bandwidth provides a larger input noise power difference compared with a fixed output referred noise, but again only to the point where the ΔG/G dominates. The third and extremely obvious option is to directly lower the receiver noise temperature (directly related to noise figure), Tn which lowers NEΔT in a linear fashion with no lower bounds on resolution as both the bandwidth-time and ΔG/G terms are scaled. Looking into the previous demonstrations [5,6,7,8,9,10] it seems obvious that these teams have all done excellent work at minimizing noise temperature with all of them reporting Tn in the 1500 to 3500ºK range (noise figures from 6-10 dB). Again looking at the previous work it also seems that an upper limit of about 10-20% fractional bandwidth seems the best achievable in silicon receiver technology. Given that the attainable Tn and B probably cannot improve much beyond these reported ranges, and the integration time, τ is usually set or at least limited by the observation scenario (especially when looking at time limited events like people standing still in front of a passive imager), the next optimization point turns to the somewhat more complicated ΔG/G term.

3. ΔG/G BEHAVIOR IN RADIOMETRY SYSTEMS The ΔG/G term of the radiometer equation refers to a quantity called by several names including “gain-variation”, “gain-fluctuation”, “gain-drift” by the astronomy and climate communities, and essentially refers to the gain of a receiver varying over time. The ΔG/G term is a statistical quantity and describes the variance of a radiometer or receiver’s gain during the integration time compared to the mean value of the gain during the same period. While circuit noises (flicker & shot) are contributors to this gain variation


Zero-MeanDominated

(moise)

Integration Time

term, any transient changes that occur on timescales shorter or comparable to the observation time affect the ΔG/G behavior. The term is often dominated by non-electrical environmental factors (thermal stability, mechanical vibration, power supply stability) making the ΔG/G behavior extremely difficult to quantify analytically, or even by using measured component parameters to calculate system behavior. Therefore the ΔG/G behavior is typically only evaluated through empirical measurements where a radiometer continuously observes a target of known temperature, and considers deviations over long periods of time. Additionally since many of the contributors of gain drift are non-zero-mean, ΔG/G itself is actually a function of τ and may actually reduce NEΔT performance if a large τ is selected as averaging a non-zero-mean process does not converge to zero. One important parameter to consider in selecting the integration time for an observation is the Allen Time, TA of a radiometric or receiver system, which describes the maximum averaging time before the contribution of the non-zero mean ΔG/G term overcomes the benefits of increased integration or averaging on reducing the zero-mean 1/Bτ term of the NEΔT expression. The entire behavior as integration time is varied for a radiometric system is typically expressed as an “Allen deviation function” (obtained through measurements) with an example shown in Fig. 1. In the short timescale regime (left hand side of the plot) the radiometer is noise dominated (1/Bτ term), and so increased observation or integration time improves the NEΔT until the TA is reached, beyond which point the non-zero ΔG/G term becomes dominant and corrupts the averaging process, lowering the radiometer’s overall NEΔT.

Fig. 1. Typical Allen deviation function showing “Allen Time” (TA), the integration time where the best NEΔT performance is obtained. The second unusual characteristic of the ΔG/G behavior is that unlike additive noise at each stage in a receiver (the type of most concern in communication systems), the effect of gain drift propagates multiplicatively instead of by superposition through cascaded stages. This means that ΔG/G contributions are equally critical in each stage of a receiver system for radiometry, and that the addition of pre-amplification will not suppress the ΔG/G contributions of later stages. Figure 2 shows a comparison of additive noise and ΔG/G behavior.


c

(.1X

W

Ant I

Y 1o_Dicke SW

50"cal"

Switch signal

LNA

LO

"sia"

"cal"

time

Vin

Vin

A(Vin + n2)

(a)

..

NF = NF1 + NF2 ...A1

(A1+ r(b)

..

) (A2 +V^n22) ..

Fig. 2. (a) Additive noise typically of concern in communication systems where the noise contribution of later stages is suppesed by the gain of preceeding stages. (b) ΔG/G behavior modeled with a random variable for the gain of each receiver stage showing how each stage’s contributions propagate multiplicatively. As the ΔG/G contributions of a receiver cannot be overcome or suppressed through low-noise pre-amplification, other forms of calibration techniques are required for a radiometer system to achieve high NEΔT performance. The most common approach for mitigating ΔG/G contributions in radiometer systems is Dicke Switching [12] which is a form of double correlated sampling, used to track and de-embed the ΔG/G behavior.

4. DICKE SWITCHED RADIOMETERS The classical approach for addressing ΔG/G behavior in radiometric systems is a method of double-

correlated sampling where a switch is placed in front of the receiver chain and selects between a load at a known temperature and the antenna input as illustrated in Fig. 3. This is often referred to as a Dicke switch (named after Robert Dicke, an astronomer who from 1940-1960 developed radio-telescopes and determined the theoretical temperature bounds of the cosmic microwave background).

Fig. 3. Dicke switched radiometer where a swtich is placed in front of a reciever and switches between a known temperature load and the antenna to capture gain changes to de-embed the ΔG/G behavior.


Detector + Output Amp

win rinnrinrinr

As Robert Dicke first noted in 1946 [12], adding this switch allows a radiometer to remove the effects of gain drift by measuring a known load (calibration measurement) to de-embed the changes in gain from the real measurements of the target. By rapidly interleaving the measurements with the calibration, the radiometer can theoretically generate two highly correlated data sets (provided the time constants associated with the gain drifts are much longer than the interleaving time) and allow us to separate output power changes due to gain changes from output power changes due to target temperature changes. Addition of the Dicke switch can be represented by several modifications to the radiometer equation ∆ = 2 1 + ( − ) ∆ Where the ΔG/G acquires a coefficient of (TI-TC)/TI which indicates how well matched the calibration resistor noise temperature TC and antenna input temperature TI are (or alternatively how well matched their resistances are). Additionally the noise temperature is inflated by a factor of 2X to account for the reduced observation time (assuming a 50% duty cycle of the Dicke switch). Note these adjustments assume an ideal switching element with infinitely fast settling time and no transient artifacts (overshoots or undershoots). Several exciting silicon radiometers [5,6,7,8,9] already implement Dicke switching although their performance and characteristics in actual radiometric observations have not been quantified in the above references. Therefore, to perform this evaluation we reuse the 65nm CMOS radiometer chip previously reported in reference [7]. Since the previous paper was published, we have re-fabricated the chip and added several digital control functions including on-chip DACs and a SPI controller interface to provide biasing voltages for easier testing, however the core mm-wave and detector circuitry remains unchanged from the prior work we’ve reported. The updated chip photo is shown in Fig. 4. While the design was only characterized with on-wafer probe measurements in [7], we have since packaged the chip on a PCB module with an antenna and microcontroller to implement the entire radiometer system, enabling direct evaluation of NEΔT performance in real-world radiometric observations.

Fig 4. Updated version of the radiometer chip previously reported in [7] and used in this work for direct NEΔT evaluations.


PatchAntenne

DICKESWITCH

A

CMOS SoC

--I accumulatorto PC

4 -I accumulator

Control ASIC 1

top control

A block diagram of the complete Dicke radiometer module is shown in Fig 5. At the core of the radiometer is the CMOS chip implemented in 65nm CMOS technology which contains the receiver chain, power detection, baseband amplification and additional control circuitry to provide digital bias trimming for the mm-wave and detector circuitry.

Fig 5. Block diagram of Dicke-switched radiometer module with CMOS SoC based receiver and power detection with external microcontroller to provide digitization/accumulation.

Within the CMOS chip each stage of the low-noise amplifier (LNA) is digitally set via a pair of R2R DACs controlled by a central SPI interface (which links to a PC). Outside the chip, a microcontroller co-located on the same PCB provides the ADC required to digitize the power detector output, provide the accumulation/readout functions and the control for the Dicke switch. The radiometer module’s antenna is implemented as a simple patch antenna fabricated on a Roger’s substrate and ribbon-bonded to the CMOS chip. It’s essentially a textbook design simply to facilitate hot and cold load (Y-factor) testing with liquid nitrogen (LN2) soaked and room temperature absorbers. Fig 6 show a close up photograph of the micro-assembly. Additionally the PCB provides all bias voltages regulated from the 5V carried on the USB interface which is used for data readout and SPI control operations.


`..

1: 13x111115 ',In

,

.'.fM

ff afé:'ó

atA

PI=

3 sioa

,..C

S,"..E

CJ

«1

:.E

Ol'

liM

^ Ti

_ reE

]5

"our»au

1Ir.

.C

Eçir C

1 :en

E1,11"r°

w'¡ ñ124 .

Fig 6. Photograph of overall Dicke-radiometer module assembly and micro-photograph of antenna and SoC chip assembly.

As CMOS flicker noise is quite high compared with III-V technologies, its contribution to ΔG/G behavior cannot be ignored and so switching beyond the flicker corner of 65nm CMOS (approximately 0.3 MHz) is required. In our radiometer this operation is implemented in software running on the microcontroller by the code shown in Fig 6. Fig 7. Coding section used to operate the Dicke switch and provide the necessary correlated double sampling to remove the effects of gain drift from the radiometer.

calibration_accum=0; measure_accum = 0; for (counter=1;counter

The parameters of integration_count can be varied to modify the integration time, but only to the point where the two accumulators (in this case 32 bit integers) begin to overflow. One subtlety here that may be a major source of confusion is the integration or observation time of the Dicke-switched radiometer is a different quantity than the integration time discussed in section II. While the radiometer integration or observation time discussed in relation to the radiometer equation is bound by the Allen time due to the ΔG/G term, the Dicke radiometer is expected to suppress this term, allowing the overall observation time to become essentially unbound. In other words the Allen deviation function of the ideal Dicke radiometer is expected to be monotonically decreasing allowing for ever improving NEΔT performance as the integration time is increased. This of course is not true in practice, as the switch and other Dicke-related electronics themselves also exhibit time-varying and non-zero mean sensitivities like the receiver chain does. For this evaluation we use 2000 cycles for our integration_count value. In this system the calibration and measurement of the radiometer are interleaved at 500 kHz. The 500 kHz measurement rate represents 1/20th of the 10 MHz clock used to drive our microcontroller. This 1/20 or 0.05 factor is applied to the full clock frequency because the microcontroller executes a total of 20 assembler instructions for each cycle of the Dicke switch operation, meaning 2000 counts at 500 KHz represents 4 ms of total integration time for combined observation and calibration. For NEΔT calculations the Dicke radiometer has a 50% duty-cycle so the effective observation time will be only 2 ms (half of the total observation time). Figure 8 shows the schematic of the Dicke-radiometer RF/analog front end (previously reported in [7]), which includes an input balun, differential LNA and power detector with on-chip Dicke switch implemented as a shunt NMOS to ground. To reduce insertion loss, the Dicke switch structure with no transistor along the signal path is adopted as shown. A fully differential LNA architecture is employed to suppress common mode and supply ripple/noise at the cost of extra power consumption. The input balun serves as the conversion from single-ended to differential signals as well as the input matching network. Figure 8 also sketches the LNA schematic, which features four-stage amplification and adopts fully differential, common source cascode structure. Transformer based inter-stage coupling and matching are utilized for compact implementation. Cascode structures provide the benefits of higher amplifier stability through input and output isolation. To alleviate stray capacitance at the cascode internal nodes, series transmission lines are inserted to tune out the parasitic capacitance. To facilitate non-coherent power detection, a differential detector structure is adopted, with the schematic shown in Figure 8. Differential detectors are known to offer better common-mode rejection and coupling noise suppression related to the supply and ground nodes.

[ ]2

LNA Detector Filter PGABalun

Front End Circuit

Dicke SW

LNA

Differential Detector

Dicke SW

Fig 8. The structure of the Dicke-radiometer front end circuits and the corresponding LNA and detector schematic


_I 1.68

Q.

8 1.67'

oE.2 1.66

5 165

1 64

1.63

Cold Load Applied® (20 seconds)

\I-verage

J

vel 1.675

Cold Level 1.629

1.62e5 10 15 20 25 30 35 40 45

Time (seconds)

5. MEASUREMENT OF THE DICKE RADIOMETER

In order to characterize the Dicke-radiometer, we first performed a series of hot and cold load tests using liquid nitrogen LN2 and absorber material. In the first test, the Dicke switch was forced to the on position and the absorber soaked in LN2 was placed in front of the antenna after the radiometer’s output had already been recorded for 20 seconds. After the LN2 is introduced the output is recorded for 20 more seconds to provide a clear baseline output noise power. The LN2 quickly evaporates returning the input temperature to the room temp of 293ºK. Fig. 9 shows the plotted radiometer output during this test (with the LN2 introduced at t=20 seconds). Using this measurement and noting the contrast between the liquid nitrogen (boils at 77ºK) and the background level (293ºK) we can directly compute the Y-factors to provide an estimate of noise performance.

Fig 9. Output of the radiometer during a 40 second test when an absorber soaked in liquid N2 is placed near the antenna at t=20 seconds and the Dicke switch remains at the antenna position.

By first taking the mean of the digital output codes from the accumulator when the nitrogen is not applied (1.675x105) then looking at the digital code when the cold load was applied (1.629x105), and understanding that this difference represents (293 – 77 ºK) of temperature contrast, we can extrapolate the total system temperature of the radiometer (including antenna, RF interfaces, and feedlines) to a value of 7650ºK. Using the IEEE specified temperature for noise factor/figure (23ºC) we can compute the overall noise figure of the radiometer system to be 14.4 dB (includes antenna, RF interfaces and feedline). From reference [7], probe-based on-wafer measurements with a W-band noise source indicated the LNA alone was on the order of a 10.5 dB NF. Second, to demonstrate the entire radiometer operation including the Dicke switch, we again perform a 40 second measurement without changing the input temperature as shown in Fig 10. The first and most obvious issue is that the output levels are very different for the calibration and measurement


1400

1200-

1000

80&

600

400

200

00

RMS. level ..(244)

Cold load Full scale (1372)

Measured NEDT= (244/1372) X 216 °K= 38 °K

10 15 20 25Time (Seconds)

30 35

s. 5ko.

oLtWrtwEoEs.cc -5k0

-Antenna- Reference

5 10 15 20 25 30 35 40 45Time (Seconds)

-Antennaa

10 15 20 25Time (Seconds)

30 35 40 45

side of the Dicke switch, suggesting there are differences between the two poles of the Dicke switch. If we assume the mismatch is a simple gain error, it can be directly corrected in software by first normalizing both accumulators, and then scaling by the ratio of the mean values for each channel as shown in Fig 11. Similarly an offset can be added and the combination of gain/offset can be optimized to provide the best possible correlation.

Fig 10. Output of the radiometer during a 40 second test when a constant temperature is applied to the input antenna.

Fig 11. Adjustment to calibrate for gain/offset mismatch between calibration and measurement data of the Dicke switch radiometer.

Third, to characterize the overall NEΔT we repeat the test and again introduce the LN2 soaked absorber at t=20 seconds with the corrections mentioned above applied and record both the antenna and reference signals, again over a 40 second period.

Fig 12. Final output of radiometer when correlated-double sampling is applied. RMS gain drift level (not counting peak) is shown along with N2 level.

c_mean = average(cal_accum); m_mean = average(meas_accum); scale = m_mean / c_mean; final_output = (meas_accum)/m_mean – scale*(cal_accum/c_mean) + offset;


P NA-X

+Trig

EXT H

50

Dicke Receiver

LNA PGA

Prog Delay I Microcontroller I

The results of this complete test are shown in Fig 12 where the difference between calibration and antenna signal is taken to provide the final calibrated radiometer output. By considering the peak response due to the LN2 soaked load and remembering that this contrast is again the difference between 293ºK (the room) and 77ºK, we can consider the RMS error due to gain drift, and directly compute our resolvable temperature difference (or NEΔT), measured to be 38ºK. Going back to the radiometer equation, and first assuming that the ΔG/G term is well suppressed, we see that with the 2 ms integration time, a noise temperature 7650ºK and our best estimate of detector 3 dB bandwidth at 10.5 GHz based on measurements from [7] that an NEΔT of no worse than 3.3ºK was expected. This is a considerable discrepancy from our 38ºK measurement so clearly our assumptions are incorrect. Given that the set integration time is fixed, the noise temperature is known with relatively high confidence, and the 3 dB bandwidth is unlikely to change from prior measured values, we must deduce that the ΔG/G term is the root cause of the discrepancy. By re-visiting the radiometer equation once more and equating the measured NEΔT to these values we can compute the actual ΔG/G term. This exercise shows a 0.0025 value for the ΔG/G term including the (TA-TC)/TA coefficient (representing a 0.0108 dB change in gain) is enough to explain the poor NEΔT results obtained in measurement. A wide range of issues related to the Dicke switch could explain this limited ΔG/G suppression including poor isolation between the two switch poles, or the receiver input matching conditions being different enough at each pole that the ΔG/G behavior is no longer well correlated between the two, limiting the correlated double sampling mechanism. Similar to overall ΔG/G related behavior, these potential Dicke switching issues are difficult to quantify analytically, and so we instead have performed an additional time-domain experiment to illustrate the potential for difficulties with Dicke-switching radiometers built from silicon receivers.

Fig 13. Measurement setup to evaluate the transient impedance of the Dicke switching radiometer’s input port.

In this experiment we used the setup shown in Fig. 13 to evaluate the transient impedance behavior at the input port of the radiometer. First a VNA (Agilent PNA-X) with a W-band extension heads (from VDI) is set to monitor the S11 of the Dicke-radiometer’s input port using an on-wafer probe similar to testing previously performed in [7]. However instead of a frequency sweep, the VNA is set to evaluate the return loss at only a single frequency point (94 GHz), and is triggered by the same signal controlling the Dicke switch. A tunable delay is introduced between the signal controlling the Dicke switch and the signal triggering the VNA (implemented on the VNA by simply adjusting the “trigger-delay” function). By then sweeping this delay and repeating the S11 measurement, the transient impedance matching behavior of the Dicke switch s can be observed (at least at a single frequency). Fig. 14 plots the results of this test as the delay is swept from 0 to 4 us after the control signal positive edge for the Dicke switch. Although this measurement only provides direct information about the antenna pole of the switch, and not the “throw” terminal at the input of the receiver, there is considerable evidence from the waveform that the transient behavior of the switch is not symmetric. As the switch moves to the reference port (the portion of


20

-300 1 2 3 4

Time Relative to Dicke Switch Signal (us)

the waveform with approximately -1.5 dB return loss), there is a large transient overshoot, while when switching to the antenna port there is a long transient settling behavior (likely from common-mode bias nodes in the receiver having long time constants). The vastly different transient behavior would suggest that the two poles of the switch have relatively poorly matched impedance conditions relative to each other, adequately explaining the poor cancellation of the ΔG/G behavior in the NEΔT measurements.

Fig 14. Results obtained when measuring the time domain impedance matching behavior of the Dicke radiometer’s input port at 94 GHz.

6. OBSERVATIONS

As the radiometer equation elegantly describes, the NEΔT grows with the inverse square of bandwidth and linearly with ΔG/G meaning relatively minor values can lead to significant loss of temperature resolution if ΔG/G is not well controlled.


25

o

20 ...

I-15

WZ

ß 10Cß

5

o0.001 0.010 1111 0.100

Relative Gain Variation AG /G (dB)

Fig 15. Calculated NEΔT for total gain variations of 0.001 dB to 0.1 dB for a radiometric receiver with a 6.2 dB noise figure and 26 GHz of bandwidth.

For Dicke-switching to be effective, the level of mismatch between the two poles of the front-end switch needs to provide the necessary suppression of ΔG/G required to support a given NEΔT (as this is the basis of double correlated sampling). Considering the extreme case of Dicke switch mismatch requirements, let us consider a “super-amazing” silicon receiver that provides a combination of the best reported radiometer noise performance (6.2 dB with Tn = 918ºK) [6] with the widest reported bandwidth of 26 GHz from [5] and a typical integration time of 10 ms for a passive imaging application. In this case we can calculate the attainable NEΔT for a given level of ΔG/G as shown in Fig. 15. Assuming we then wanted a NEΔT no worse than 1ºK (typical of passive imaging and similar applications) would require us to achieve a ΔG/G term including the (TA-TC)/TA factor of no worse than 0.00054 with our Dicke switch. In terms of circuit design, this would suggest that the mismatch of the Dicke switch poles becomes far more important than the inflation of noise temperature due to its insertion losses.

Alternatively let’s also consider implementation of a similar system in the traditional III-V receivers used for most astronomy and weather sensing applications. In this scenario, modern InP-hemt based technology can offer receiver noise temperatures (Tn) as low as 20-30ºK [12] when cryogenically cooled to the 77ºK range (as is often done for these applications). Using these values, and remembering that Tn directly scales the contributions of both the 1/Bτ term and ΔG/G to NEΔT we see that the matching and ΔG/G suppression requirements of a Dicke switch in these technologies is much relaxed. For example, to maintain the same NEΔT of 1ºK with the same 26 GHz bandwidth as our “super-amazing” silicon receiver with the same integration time of 10ms, the III-V device requires a final ΔG/G including the (TA-TC)/TA factor of only 0.017 which is more than 2 orders of magnitude better than the situation in silicon technology.

7. CONCLUSIONS

Although this evaluation is mostly empirical in nature due to the difficulties of estimating ΔG/G behavior analytically, we can make two key conclusions for the continued development of silicon radiometer systems:


First, it is obvious from this exercise that the RF receiver parameters alone: Bandwidth, Noise Figure (or Tn) and Integration Time do not adequately describe the attainable NEΔT from a complete radiometer system as a small ΔG/G contribution can greatly limit the resolution, and the presence of a Dicke switch does not imply complete cancellation, or even imply cancellation to acceptable levels of of ΔG/G will be achieved. Second, while continued improvement of noise figure (or noise temperature) and wider bandwidth receivers are important to enable high performance silicon radiometers, equally, and possibly even more critical is the continued development of extremely well matched and well behaved RF switches for future Dicke-switched radiometry.

ACKNOWLEDGEMENTS

The authors would like to acknowledge TSMC for their excellent 65nm foundry support as well Jacob Kooi of JPL for several excellent technical discussions related to the processing of NEΔT measurements.

REFERENCES

[1] D. R. Easterling, G. A. Meehl, C. Parmesan, S. A. Changnon, T. R. Karl, and L. O. Mearns, “Climate Extremes: Observations, Modeling and Impacts,” Science 22, Vol. 289, No. 5849, pp. 2068-2074, Sep. 2000.

[2] S. Brown, B. Lambrigtsen, A. Tanner, J. Oswald, D. Dawson, and R. Denning, “Observations of tropical cyclones with a 60, 118 and 183 GHz microwave sounder,” in Proc. IGARSS, Jul. 23–28, 2007, pp. 3317–3320.

[3] A. Siegenthaler, O. Lezeaux, D. G. Feist, and N. Kämpfer, “First water vapor measurements at 183 GHz from the high alpine station Jungfraujoch,” IEEE Trans. Geosci. Remote Sens., vol. 39, no. 9, pp. 2084–2086, Sep. 2001.

[4] S. Gulkis, et al., “Remote sensing of a comet at millimeter and submillimeter wavelengths from an orbiting spacecraft,” Planet. Space Sci. 55, 1050-1057, 2007.

[5] Leland Gilreath, et-al, "Design and Analysis of a W-Band SiGe Direct-Detection-Based Passive Imaging Receiver", IEEE Journal of Solid-State Circuits, Vol 46, No 10, Oct 2011.

[6] J. May and G. Rebeiz, "Design and Charact. of W-Band SiGe RFICS for Passive Millimeter-Wave Imaging", IEEE Trans. MTT, Vol 58, No 5, pp 1420-1430, May 2010.

[7] Qun Jane Gu, Zhiwei Xu, Heng-Yu Jian, M. F. Chang, "A CMOS Fully Differential W-Band Passive Imager with

Documents

PROCEEDINGS OF SPIE - ntut.edu.twwangsen/Courses/A CMOS... · design and implementation of mm-wave radiometers with excellent SiGe demonstrations at frequencies up to 100 GHz with