ELECTRO-MAGNETIC REMOTE SENSING DEFENCE TECHNOLOGY CENTRE (EMRS-DTC)

Dynamic range improvements and measurements in radar systems

B.J. Harker, Z. Dobrosavljevic, E.P. Craney, C.M. Tubb and G.L. Harris Abstract: The extension of the dynamic range performance of existing radar systems is integral to the requirement for current sensors to see deeper into clutter, that is, to detect weak signals (e.g. small targets) in the presence of strong interference (e.g. urban clutter signals). The rationale behind this requirement is that this additional capability greatly enhances the utility of military and commercial systems. Two approaches to improving the dynamic range of modern radar and surveillance receivers are described: (1) linearisation of intermodulation distortion (IMD) using digital post-distortion algorithms and (2) linearisation of IMD using frequency retranslation mixer (FRM). The experimental results presented illustrate that the digital post-distortion and FRM approaches are effective in providing IMD improvements for the hardware and test scenarios measured. The improvements measured ranged from 15 to 35 dB for narrowband signals to 2 6 dB for frequency-modulated continuous wave (FMCW) wideband signals. The digital post-distortion approach provided the greatest performance improvement when the input signals were between the analogue-to-digital converter full range and 20 dB below that level. The improvements measured for the FRM linearisation approach ranged from 20 to 25 dB for the narrowband two-tone tests and 16 18 dB for the wideband FMCW signal chirps.

Introduction

Military and commercial electronic sensors and communications systems are limited by the dynamic range of the hardware used to implement them. In radar, a problem exists when trying to detect small targets in a highly cluttered environment. In particular, urban clutter and jamming can quickly drive the receiver front-end low noise ampliers, mixers and analogue-to-digital converter (ADC) into their nonlinear operating region and thereby suppress sensitivity to small targets. In communication receiver systems, it is still practically impossible to operate a receiver at its full sensitivity near to base stations and other transmitters (both in and out of band). The classic solution is to use automatic gain control (AGC) or sensitivity time control. Unfortunately, this produces changes in receiver sensitivity which the operator may not be aware of meaning they cannot manually compensate for the changes. This research work, which was undertaken for the Electro-Magnetic Remote Sensing (EMRS) Defence Technology Centre (DTC) [1], endeavoured to take a holistic view of this problem in order to provide methods and techniques to assist across a number of domains. The novel element to the work has been the expansion and application of previously studied techniques to new radar test-bed systems, so as to demonstrate and quantify the performance improvement which may be achieved. Two techniques have shown particular promise over the course of the programme, sophisticated digital linearisation
# The Institution of Engineering and Technology 2007 doi:10.1049/iet-rsn:20060175 Paper rst received 14th December 2006 and in revised form 14th July 2007 The authors are with Roke Manor Research Limited, Roke Manor, Romsey, Hampshire SO51 0ZN, UK E-mail: brett.harker@roke.co.uk

and mixer linearisation. This paper outlines the techniques and describes the results of applying these techniques to real radar hardware. 2 Markets and applications

Applications for dynamic range enhancement methods include complex electronic sensors such as [2 4]: advanced multi-function radar (MFR); phase array radar; electronic surveillance measures; advanced bistatic radar; active microwave/millimetre-wave imagers.

Other applications include radio/wireless communications systems such as [2 4]: cellular systems, mobile telephone base stations; 3G (UMTS), 4G (OFDM-MIMO), ultra-wide band (UWB) systems. 3 3.1 Digital linearisation Concept of post-distortion

The function of digital post-distortion is to improve the linearity of a receiver and hence the dynamic range of the radar receiver. The improved linearity will be observed as a reduction in the power of intermodulation distortion (IMD) products and spurs caused by strong signals present at the receiver input. Using this approach, embedded digital linearisation should improve the spurious-free dynamic range (SFDR) of the radar receiver. In a linear receiver, ignoring signal down-conversion, and representing signals in their discrete form, output of the
IET Radar Sonar Navig., 2007, 1, (6), pp. 398 406

398

receiver will be a linear transform, that is, convolution of the processed signal x[n] and the receivers transfer function h[k] X y[n] h[k ] x[n k ] (1)
k

did not exist in the received signal y(t) y1 (t) y2 (t) IM(t, x(t)) (5)

where n represents time and k represents the delay. If the receiver is linear, a sum of strong jammer x1[n] and a weak desired signal x2[n] at the receiver input will result in an equivalent sum of two signals at the receiver output X y[n] h[k ](x1 [n k ] x2 [n k ])
k

y1 [n] y2 [n]

(2)

Since the output is a sum of two independent signals, the wanted signal y2[n] can in principle be separated from the unwanted signal y1[n], for example, using a correlation technique. In a nonlinear, distorting receiver, its transfer function h[k] becomes a function of input signal, as well as the delay, h[k, x[n]] X y[n] h[k , x[n]](x1 [n k ] x2 [n k ])
k

where IM(.) are the intermodulation products and x(t ) is the input signal x1(t ) x2(t ). These products are the function of time and both input signals. The intermodulation products can be strong enough to mask the weaker desired signal. This scenario is illustrated in the following diagram. Block diagram in Fig. 1a shows a simplied receiver with an analogue front end that is distorting due to the presence of strong unwanted signal (i.e. jammer) at the receiver input. The jammer-to-signal ratio at the receiver input is greater than the SFDR. As a result, the products of distortion, marked as IM(.) in Fig. 1a, are masking the desired signal and the jamming is effective. Fig. 1b shows a corrective action, or post-distortion, implemented in the digital stage of the receiver. On the basis of the available information on the distorting characteristics of the receiver and the jammer signal, the postdistortion stage can perform a corrective action, thus suppressing the intermodulation products and improving the receiver SFDR. The output of the post-distortion stage now becomes z(t) F (y(t)) y1 (t) y2 (t) im(t, x(t) (6)

= y1 [n] y2 [n]

(3)

This dependence can be expressed as a Volterra series X y[n] h1 [k ]x[n k ]

k

X
k1 ,k2

h2 [k1 , k2 ]x[n k1 ]x[n k2 ] X

k1 ,k2 ,..., km

h m [ k1 , k2 , . . . , km ] (4)

x[n k1 ] x[n km ]

where hi[k1 , k2 , . . . , km] are Volterra kernels of the ith order. Using this notation, the higher-order (i . 1) Volterra kernels will introduce a component into the output that

where F(.) here marks the transfer function of the postdistortion stage and im(.) marks the attenuated intermodulation products. If the IM products are sufciently suppressed, the desired signal will emerge from them, thus becoming detectable by the receiver. The post-distortion stage has to be implemented in a nonlinear form. Volterra lters are a popular architecture to provide nonlinear transfer function, and one use of such lter in distortion mitigation is described in [5] where experiments are used to demonstrate a 6 dB improvement in dynamic range for the specic hardware conguration being considered. In this paper, a different approach, where Volterra lter is used as a post-distortion stage, is described.

Fig. 1 Concept of receiver digital post-distortion

a Without post-distortion b With post-distortion
IET Radar Sonar Navig., Vol. 1, No. 6, December 2007

399

Fig. 2 Generic receiver digital post-distortion conguration and algorithm

3.2

Post-distortion algorithms

The generic receiver digital linearisation or post-distortion conguration is illustrated in Fig. 2 [4]. In this diagram, an input signal X is passed through an unknown linear model G to create an output Y G(X) with nonlinear distortion. To linearise the nonlinear output Y G(X), an inverse nonlinear model estimate H is calculated as shown in Fig. 2. During adaptation, the post-distortion lter coef^ Y forms the input to the cients will be unity such that Y adaptive model. The coefcients of the inverse model H can then be adapted to minimise some function of the ^ . Normally, the mean squared error signal E X 2 X error criterion is used for adaptation purposes. Once the adaptive model has converged, the coefcients of the non-linear lter model H are copied to the post^ H(Y), distortion lter to create a linearised output Y which will ideally match the original input signal X with no nonlinear distortion (in the mean squared sense). For the post-distortion application, the signal Y would be original receiver ADC output with nonlinear distortions, and the signal X would be some known ideal reference or training signal corresponding to Y, but with no nonlinear distortion. One of the most widely used forms of nonlinear inverse lter models is the Volterra series model. The discrete

Volterra series post-distortion model for a (2p 2 1)th order bandpass nonlinearity is given by the following equation
(1) M 1 X

^n x

^ ni1 h(1) i1 y

(3) (3) (3) M 1 M 1 M 1 X X X

i1 0

i1 0

i2 0

i3 0

^ ni1 y ^ ni2 y ^ h(3) ni3 i1 ,i2 ,i3 y

p1) M (2 X1

p1) M (2 X1

(7)

p1) ^ hi(2 y 1 ,..., i2p1 ni1

i1 0

i2p1 0

^ nip y ^ ^ y nip1 y ni2p1 The likely bandpass nature of the nonlinearity G(X) in the post-distortion application means that only odd-order terms are included in the Volterra series; all even-order terms (and some odd-order terms, e.g. terms of the form v1 2v2) would lie outside the frequency band of interest. The complexity of the higher-order terms means that normally only the rst and third-order terms are considered. In the following sections of this paper, we describe the experimental tests and results obtained following application of the digital receiver post-distortion method to Rokes bistatic radar and Rokes MFR test-bed. The

Fig. 3 Bistatic radar wideband signal test set-up

400
IET Radar Sonar Navig., Vol. 1, No. 6, December 2007

Fig. 4 Rokes bistatic radar CONOPS (equivalent to wideband signal test conguration)

experiments were completed using typical narrowband and wideband test signals. 3.3 Rokes bistatic radar

3.3.1 Test-bed: The rst radar sensor test-bed used during verication testing of the post-distortion algorithms consisted of Rokes bistatic radar system [4]. 3.3.2 Experimental conguration: Fig. 3 illustrate the bistatic radar test conguration used during wideband verication testing of the digital post-distortion algorithms [4]. A similar set-up was used during narrowband signal testing. The test signals shown in Fig. 3 are equivalent to the actual signals shown in the Concept of Operations (CONOPS) diagram, which is illustrated in Fig. 4. 3.3.3 Experimental results: The linearisation curves plotted in Fig. 5 illustrate the experimental results for the bistatic radar test results for both narrowband (two-tone) and wideband test signals. The wideband test signals used were modulated to give bandwidths of less than 1 2 MHz and the generated waveform occupied most of the receiver channel bandwidth, and the intermodulation products therefore mostly overlapped in frequency with the test signals [4]. The algorithm was trained with the peak signal amplitude reaching the full ADC dynamic range. However, as the training signal was a non-constant amplitude signal, samples of the training signal were distributed throughout the ADC range, from smallest to largest values.

The algorithm used in receiver post-distortion had a memory length of one (i.e. the delay line had two taps) and was of the seventh order. This meant that IM products of up to seventh order could potentially be suppressed. Fig. 5a shows the suppression of IMD products for the different orders as a function of the signal level below the receiver saturation point. Referring to the linearisation plots in Fig. 5a, the abscissa (x-axis) is the level of input signal relative to the full range of the ADC in the receiver. The ordinate ( y-axis) is the ratio of IMD products before and after post-distortion, that is, the amount of IMD suppression or improvement. During the experiments, the amplitude of the test signals, relative to the ADC clipping level were varied by 3, 6, 10 and 20 dB. It can be seen from Fig. 5a that the suppression of IMD products or improvement for full-scale input signals is of the order of 1116 dB. As the input signal level is reduced, the receiver becomes more linear, and the improvement decreases. With the input signal level at 20 dB below the full-scale input signal level, intermodulation products generated in the receiver became too weak to be detected; therefore the improvement due to postdistortion could not be measured and has been interpreted as equal zero. Fig. 5b illustrates the measured suppression of wideband intermodulation products, which is of the order of 0.5 to 2 3 dB. The measurement included IM Noise and the random noise power component degrades the accuracy of the measurement, meaning actual improvement in practice will be better than that measured.

Fig. 5 Linearisation curves for Rokes bistatic radar showing IMD improvement
a Using narrowband signals b Using wideband signals
IET Radar Sonar Navig., Vol. 1, No. 6, December 2007

401

Fig. 6 Rokes MFR test-bed wideband signal set-up

3.4

Rokes advanced MFR test-bed

3.4.1 Test-bed: The second radar sensor test-bed used during verication testing of the post-distortion algorithms consisted of Rokes MFR system [4]. 3.4.2 Experimental conguration: Fig. 6 illustrates the MFR test-bed conguration used during wideband verication testing of the digital post-distortion algorithms [4]. The wideband signal used in the tests was a combination of frequency-modulated continuous wave (FMCW) chirp signals simultaneously sweeping through different parts of the receive band. A similar set-up was used during narrowband signal testing. 3.4.3 Experimental results: The linearisation curve plotted in Fig. 7 illustrates the MFR test results for the narrowband standard two-tone test signal case [4]. In Fig. 7, the IMD improvement is plotted as a function of the input signal power level for narrowband signals. The plotted IMD improvement is given for the important third-order products. From Fig. 7, it can be seen that the strongest third-order intermodulation (IM3) products are being suppressed by  35 dB maximum. Fig. 8 illustrates the MFR test results for an FMCW chirp test signals [4]. All IMD products clearly still fall in-band and are mainly concentrated in the adjacent frequencies at the edges of the chirped signal bandwidth. The reduction of IMD products in the adjacent channels can be observed. The strongest IMD products are being suppressed by  5 6 dB. An alternative view of the IM terms and the resulting post-distortion IMD improvements are shown in the spectrograms illustrated in Fig. 9. Fig. 9a shows the test results before post-distortion algorithm is enabled. Fig. 9b

illustrates the spectrogram for the corrected composite chirp after the digital post-distortion algorithm is enabled. The white (light-coloured) traces on the spectrograms are higher-order IMD products. The reduction in white (lightcoloured) traces in Fig. 9b shows that the post-distortion improvement is effective. It may be observed from the experimental results that the IMD improvements for the narrowband tests are signicantly better when compared with typical wideband tests. For the narrowband two-tone test signals, the IMD improvement is relatively high because the adaptive linearisation algorithms can fully converge and hence accurately compute and compensate for the nonlinear distortions introduced by the receiver. The adaptive linearisation does converge, but not fully. This means that IMD improvement achieved for wideband signals is less than that achieved for narrowband signals. 3.5 Variations in experimental results

The experimental results for the bistatic radar system (Section 3.3.3) and the MFR test-bed system (Section 3.4.3) are clearly different when compared. This section of the paper explains the reasons for the differences in SFDR performance improvements for the two systems investigated. In a post-distortion approach to receiver dynamic range improvement, performance of the correction stage depends on two important aspects. Receiver analogue front-end linearity. The less linear this analogue front end is, the more intermodulation products it will generate. As such, poor performance receivers will have small SFDR before post-distortion, giving more room for potential linearity improvement using postdistortion, and therefore the post-distortion will show more improvement. In a good receiver that is very linear and has a relatively high value SFDR to start with, there are limited opportunities for additional improvement of linearity, and such receiver post-distortion technique will give less SFDR improvement. Input signal. Power and spectral content of IMD products depends on, for example, input signal power, peak to mean ratio and bandwidth. SFDR improvement dependence on input signal and receiver front-end characteristic was one of the reasons for undertaking measurements on different hardware platforms and with different test signals. 4 4.1 Mixer linearisation Distortion

Fig. 7 IMD improvement achieved with Rokes MFR test-bed for narrowband signals
402

The adaptive digital post-distortion algorithms described in this paper have certain limitations. One limitation is that they cannot cancel distortion originating from signals not
IET Radar Sonar Navig., Vol. 1, No. 6, December 2007

Fig. 8 Rokes MFR test-bed wideband signal test results showing IMD improvement

present in the training data. This is of primary concern in relation to mixing products where out-of-band signals generally exist during the rst mixing stage and these produce in-band distortion products. In this instance, the out-of-band signal are normally ltered in later stages meaning their in-band distortion affects cannot be replicated and hence removed by the digital correction algorithms. In this scenario, methods for mixer circuit linearisation are needed to correct for IMD introduced by the rst mixing stage. 4.2 Frequency retranslation method

The frequency retranslation mixer (FRM) method works on the principle of pre-distorting the input to the main mixer [6]. In the FRM method, the ideal pre-distorted signal is of equal magnitude but in anti-phase with the distortion introduced by the mixer stage. Hence, at the output of the mixer after down conversion, the nonlinear distortion is cancelled. The architecture of the frequency retranslation scheme generates the pre-distortion via a feedback loop by rst up converting the distorted intermediate frequency (IF) signal. It then adds, in anti-phase, a scaled version of the input radio frequency (RF) signal so as to generate an

error signal containing only the fed back nonlinearity. This measure of distortion is then fed back after scaling and phase adjustment. Fig. 10 shows the architecture and processing of a two-tone RF input signal. As explained in [4, 6], exact phase and gain matching are required to obtain the best intermodulation performance. It has been shown [6], based on the theory of vector cancellation, that 0.1 dB gain and 18 phase match are required for around 30 dB of vector cancelling. Therefore the FRM method will require similar level of matching in order to achieve a similar level of IMD suppression performance. Hence, the key to getting the technique to work effectively is obtaining good gain and phase matching. However, the theory also indicates that less perfect matching can provide vector suppression, albeit with lower performance. In order to achieve good gain and phase matching, an adaptive digital compensation approach is needed to accurately control the FRM components. The compensation approach selected was based on a high performance digital signal processor (DSP)/eld-programmable gate array (FPGA) platform. The following two approaches to realising the embedded adaptive compensation algorithms were studied.

Fig. 9 Rokes MFR wideband signal test results showing spectrogram of IMD improvement
IET Radar Sonar Navig., Vol. 1, No. 6, December 2007

403

Fig. 10 Architecture for FRM

A fully adaptive system that minimises the system error via the steepest-descent method. A lookup table method using training sequences. The experimental results presented in Section 4.4 show that it is possible to obtain stable intermodulation suppression for relatively large bandwidth signals. The results also show that the phase and gain settings give consistent performance, that is, the same IM suppression is observed, with xed gain and phase settings, for the same RF input signals. However, for a practical system, an embedded digital adaptive control algorithm is necessary to provide good gain and phase matching in order to achieve the required performance. For example, the optimum system parameters will change with input power and frequency and mixer circuit component degradation will cause slow drift in optimum performance settings. The compensation approach investigated was based on a high performance DSP/FPGA platform. The following two approaches to realising the embedded adaptive compensation algorithms were studied. A lookup table method using training sequences. A fully adaptive system that minimises the system error via the steepest-descent method. The consistency of wideband performance for a xed set of parameters indicates that the lookup table architecture could allow good overall performance. The advantage of

then making the lookup table adaptive would help counteract slow changes in the mixer circuit over time and also aid in the training procedure for new mixer circuits. It is felt that the steepest-descent-based algorithm would give the major advantage that it would continuously be seeking the optimum parameters for given input signals and mixer circuit properties. This additional performance, however, would come at the cost of greater computational complexity. The two methods both show promise in terms of controlling the retranslation circuit and it is felt that the level of parameter control required and system stability will dictate which system algorithm will give the best performance. 4.3 Test-bed and experimental conguration

Fig. 11a illustrates the FRM test-bed studied, designed and built for verication testing of the mixer linearisation method and algorithms [4]. The FRM test-bed shown in Fig. 11a is based on the architecture described in Fig. 10. The test-bed and the test equipment used during the experiments are shown with annotations in Fig. 11b [4]. In Fig. 11a, DSP control of the system was provided by an advanced DSP/FPGA signal processing board. This combined a powerful DSP and a useful set of onboard peripherals. The 600 MHz dual core processor provided a high level of processing power when prototyping the system.

Fig. 11 Rokes FRM test-bed and experimental set-up

a Rokes FRM test-bed b Rokes FRM test-bed experimental set-up
404
IET Radar Sonar Navig., Vol. 1, No. 6, December 2007

Fig. 12 FRM narrowband and wideband signal test set-up

a FRM test results for narrowband signal test set-up b FRM test results for wideband signal test set-up

Advanced vector modulators were used to give the phase and gain control for the frequency retranslation algorithm. These operate within a frequency range of 1.5 2.4 GHz and provide full 3608 phase control and magnitude control over a range of 30 dB. In Fig. 11b, the two RF inputs were driven via test signal generators and both the test mixer and IF output were monitored using Spectrum Analyser measurement equipment. The local oscillator (LO) for all three mixers was derived from a low noise signal generator, where the output level was set at 13 dBm. 4.4 Experimental results

Similar tests were completed using 10 MHz bandwidth test signals as inputs to the FRM experimental test-bed set-up. In the 10 MHz signal bandwidth experiment, the important upper intermodulation products were cancelled by  10 dB, with the less important lower products cancelled by 2 3 dB. The experimental test results presented here indicate the practical advantage of the FRM technique for detecting weak echo signals and providing dynamic range enhancement in radar receivers. 5 Conclusions

Fig. 12a illustrates the test results using narrowband signals both with and without FRM circuit enabled [4]. The results demonstrate that the method provides a signicant reduction in IM3 products of  20 25 dB. The IMD improvement for fth and seventh order products is  5 dB. The FRM test-bed conguration was optimised for IM3 performance improvements because this type of distortion is generally the most problematic in receivers. In Fig. 12a, the in-balance in the IM3 products can be put down to phase differences in the feedback path between the two IM3 frequencies. Hence, the current gain settings give imperfect phase matching for the lower IM3 product and good phase matching for the higher IM3 product. Fig. 12b illustrates the test results using an FMCW wideband signals both before and after the FRM technique is enabled [4]. The IMD improvement is  16 18 dB. In the test scenario of Fig. 12b, a low-level wanted target signal is hidden by the IMD products. Once the FRM is enabled, the IMD is reduced signicantly, meaning the wanted signal can be reliably detected. The result demonstrates the practical dynamic range advantage of the FRM method when used to detect the low-level target signals masked by IMD. In Fig. 12b, it can be seen that there is some minimal insertion loss (IL) due to the additional circuitry associated with the FRM method. However, the overall dynamic range for detecting the weak signal has been markedly increased with the application of the FRM technique. Clearly, there is a trade-off to be made between the IL for the FRM circuit and SDFR performance improvement.
IET Radar Sonar Navig., Vol. 1, No. 6, December 2007

This paper has described two approaches to increase the dynamic range of modern radar and surveillance receivers: (1) digital lineariser using receiver post-distortion and (2) mixer linearisation using FRM. The FRM linearisation has been shown to be effective in providing  20 25 dB IMD improvement for narrowband tests using two-tones and 16 18 dB IMD improvement for wideband signal experiments using FMCW chirps. The digital post-distortion approach have shown to be effective in providing IMD improvements ranging from 15 35 dB for narrowband signals (Figs. 5 and 7) to 2 6 dB for FMCW wideband signals (Figs. 5 and 8). IMD linearisation curves have been created for Rokes advanced MFR test-bed and bistatic radar system. The digital post-distortion algorithms provide the greatest performance improvement when the input signals are in the range from 1 dB CP to 20 dB below CP. For input signals 20 dB below CP, the receiver is effectively operating in its linear region, meaning the IMD improvement provided by the method is not applicable. The two techniques described in this paper are intended to mitigate different types of receiver distortion. Therefore the described techniques may be used separately or in combination to potentially improve the performance of a system. An element of the next phase of work would aim to verify through experiments that the two techniques are complementary and quantify the system performance improvement feasible by applying both techniques in combination to the systems under consideration. In implementing the receiver linearisation methods described in this paper, there are relatively signicant
405

dynamic range enhancements or IMD improvements to be gained at the system level. Nonetheless, the algorithms used to realise the methods appear computationally intensive/complex and may therefore require high performance DSP/FPGA platforms to realise their full performance potential. This cost must be weighed against the costs of developing a higher dynamic range receiver, which in some cases may move a design out of the realm of low cost COTS components and thereby signicantly impact cost and design time. It is interesting to note that in transmitter linearisation similar complexity arguments were made but today such linearisation approaches are the norm. The results conrm that accurate fully adaptive models have been created for Rokes bistatic radar and Rokes advanced MFR test-bed. The research results show that the algorithms are effective in correcting for receiver nonlinearity or distortion affects. In summary, adaptive correction methods have been studied and algorithms have been verication tested using real radar receivers. A Patent Application has been led to protect the subject matter of this work.

Acknowledgments

The research work reported in this paper was sponsored by the EMRS DTC, established by the UK Ministry of Defence (UK MoD) and run by a consortium of BAE Systems Avionics, Roke Manor Research, Thales Defence and Filtronic. 7 References

1 http://www.emrsdtc.com/, accessed December 2006 2 Pennington, D.: Dynamic range improvements in radar systems using transmitter linearisation. Roke Manor Research Ltd, EMRS DTC 1st Annual Technical Conf., May 2004, p. C15 3 Harker, B.: Dynamic range improvements in radar systems. Roke Manor Research Ltd, EMRS DTC 2nd Annual Technical Conf., June 2005, p. A10 4 Harker, B., Dobrosavljevic, Z., Craney, E.P., Tubb, C.M., and Harris, G.L.: Dynamic range improvements in radar systems. Roke Manor Research Ltd, EMRS DTC 3rd Annual Technical Conf., July 2006, p. A23 5 Rabideau, D.J.: Hybrid mitigation of distortion in digital arrays. Proc. IEEE Int. Radar Conf., 2005 6 Neisimoglu, T., Beach, M.A., Warr, P.A., and Macleod, J.R.: Linearised mixer using frequency retranslation, Electron. Lett., 2001, 37, (25), pp. 14931494

406

IET Radar Sonar Navig., Vol. 1, No. 6, December 2007