Sampling technology has evolved into a powerful tool for microwave engineers. Effective sampling circuitry requires knowledge of time-domain and broadband techniques, both of which have been advanced at Picosecond Pulse Labs (Boulder, CO) in support of several product lines, including pulse generators, comb generators, and high-speed sampling modules. The firm has succeeded in pushing microwave sampling technology beyond 100 GHz with the help of its innovative nonlinear-transmission-line (NLTL) technology (Fig. 1).
The history of microwave sampling extends back 50 years.1 It is essentially a simple process, in which a switch controlled by a local-oscillator (LO) strobe opens a path between the RF input port and an output intermediate-frequency (IF) port for an extremely short interval. In that short time, a portion of the RF signal is captured as the frequency-translated IF signal. Much of the effort to improve microwave sampling has involved work on the LO strobe drive and the sampling switch. The innovations at Picosecond Pulse Labs stem from work at its Oregon-based GaAs and thin-film facility (see sidebar). The GaAs fabrication facility creates high-performance air-bridged Schottky diodes with multi-THz cutoff frequencies that can be used as both varactors for millimeter- and submillimeter-wave frequency multiplication and as ultra-fast switches in mixers and samplers.
The firm's NLTLs are synthetic transmission lines, in which Schottky varactor diodes are distributed along a high-impedance transmission line. The lumped capacitances of these diodes are also distributed along the line (Fig. 2). The capacitance of a reverse-biased Schottky diode is nonlinear (voltage dependent) such that the capacitance at low reverse bias is much greater than the capacitance at high reverse bias. Because of this, the propagation velocity of a Schottky varactor based NLTL is voltage dependent. A large step signal that transitions from low to high voltage will be compressed in time as the initial low-voltage portion of the step travels down the line slower than the later, higher-voltage portion of the step. Consequently, the higher-voltage portion of the waveform "catches up" with the lower-voltage portion of the step, resulting in "edge compression," increasing the edge speed of the low-to-high transition. The high-speed diodes used in these NLTLs are capable of processing signals with subpicosecond transition times. To avoid aberrations due to shockwave formation, the NLTLs used in the company's flat-top step pulse generators are carefully tailored by "chirping" the diode size and line parameters along the length of the line.
Sampler performance depends on the qualities of the strobe pulse used to open and close the switches, the speed of the switches, and the circuit interconnections. An ideal switch has no series resistance and no parasitic capacitance. In other words, ideal switching diodes must have a very high cutoff frequency relative to the signal being sampled. The GaAs diodes fabricated in the Oregon facility exhibit cutoff frequencies in the THz range.
Even at lower bandwidths, other factors such as sampling efficiency (which affects noise), the linearity of the sampling process (which determines dynamic range), and isolation are all important parameters that critically depend on the strobe drive characteristics. Both the pulse shape and duration are key to critical real-world sampling circuits. A fast, square pulse will turn on the sampling diodes quickly, minimizing the time that the diodes operate in the nonlinear region of their transfer curve. A square strobe will also minimize input and bias dependant sampling efficiency by preventing input voltage-dependent aperture duration. The strobe must also be able to deliver enough current to drive the backshort. The reflection of the strobe drive waveform from the backshort turns off the sampling diode and determines the duration of the sampling aperture. Very narrow apertures are needed for high bandwidth, and therefore very fast edges are required for square apertures in the highest bandwidth samplers.
Traditionally, discrete step-recovery diodes (SRDs) have been used in strobe drive circuits. The fastest SRDs currently available have transition times of many tens of picoseconds. As a result, traditional samplers with 5 to 7 ps apertures (bandwidths of 50 to 70 GHz) have triangular or Gaussian apertures, and are necessarily very nonlinear since their aperture durations are input voltage dependent. Changing signal slew rates modulate the sampling aperture for Gaussian-shaped strobe pulses, resulting in dynamic distortion, while square-shaped apertures are relatively unaffected.
This nonlinearity can amount to as much as 30 percent over a modest 500-mV input range, and must be corrected in software. Dynamic nonlinearity, caused by input slew-rate aperture dependence, is not easily correctable and leads to significant intermodulation distortion (IMD) in high-bandwidth samplers. For this reason, Picosecond Pulse Labs uses NLTLs monolithically integrated with the sampling structure to produce the strobe drive waveforms, resulting in dramatic improvements in dynamic range and linearity. The company's advanced sampling technology is essential to a recent line (the Wave Expert series) of digital sampling oscilloscopes from LeCroy Corp., including one model with a sampling bandwidth of 100 GHz.
Getting the RF input signal to the sampling diodes is currently the main limitation to sampler bandwidth. Traditional approaches have involved the use of resistive-capacitive (RC) "peaking" to extend the bandwidth of a 50-GHz module to 70 GHz, at the expense of input matching. For the 70- and 100-GHz sampler modules employed in the LeCroy scopes, major improvements were obtained in the 1-mm and 1.95-mm coaxial interconnects. As outlined in US Patent No. 6,900,710, the GaAs sampler die penetrates a coaxial 1-mm airline cavity. The center conductor of the coaxial airline contacts the GaAs sampler die through a gold bump formed on the sampler die. This creates a "through" sampler, in which the coaxial signal line is passed through the sampler to be reused or terminated externally. This feature was also exploited in a prototype 100-GHz sampler module specifically designed to ride atop 1-mm wafer probes to facilitate on-wafer measurements.
Figure 3 shows a typical fall time measurement of one of the 100-GHz samplers. The time difference between the 90 and 10 percent points is 4.3 ps (assuming a 10 percent overshoot in the measured signal). This fall-time measurement includes the effects of the stimulus, the sampler and the total system jitter. Estimating an equal contribution to the fall time from the stimulus and sampler, and assuming that the jitter contribution of is negligible, a fall time of 3 ps is estimated for the sampler, corresponding to a 3-dB bandwidth of approximately 120 GHz.
The sampling rate of an equivalent time sampler traditionally must be traded off against the noise introduced by the IF charge amplifier bandwidth. In the design of the LeCroy sampling modules, the maximum sampling rate was set at 10 MHz, more than 20 times greater than competing products. This required drastic changes to the architecture of the IF charge amp using a "pulse resolved IF" technique where, instead of integrating the charge of the sample in a low-bandwidth, high-impedance opamp, the IF sample pulse is amplified in a bandpass transimpedance amplifier and resampled at the peak of the amplified pulse. This architecture eliminates the traditional trade-off between bandwidth and noise since the signal-to-noise ratio (SNR) improves with increasing bandwidth of the IF chain (contrary to conventional wisdom, which has SNR decreasing with bandwidth). This architecture has allowed the firm to raise the sampling rate to several GSamples/s, enabling "real-time" microwave sampling without increasing input-referred noise.
Since the NLTL approach provides very energetic sampling strobes, the firm's samplers support a large input signal range with excellent linearity. Where traditional samplers exhibited more than 10 percent nonlinearity over a 500-mV input range, the samplers used in the LeCroy scopes have less than 1-percent deviation from linearity over a 1-V dynamic range.
Downconverters translate RF and microwave signals to lower-frequency IFs that can be sent to an analog-to-digital converter (ADC) for further processing. Frequency mixers are commonly used for downconversion, although the function can also be performed with a sampling structure. This latter approach allows the ADC clock to serve as the system LO, greatly simplifying the RF front-end architecture. Sampling downconversion can be performed in a single step, rather than the multiple mixers and LOs needed (depending on the input frequency) in some superheterodyne receivers.
A high-bandwidth sample-and-hold (S/H) ahead of a high-speed digitizer allows "bandpass sampling" or "subsampling." Signals well above the Nyquist range of the ADC can be sampled without alias distortion as long as the bandwidth of the signal fits into the Nyquist band of the digitizer. The sampler extends the analog bandwidth of the ADC to the microwave or millimeter-wave band. In the past, sampler performance has not been good enough for many downconversion applications, particularly with respect to intermodulation distortion and dynamic range. However, the firm's advances enable significantly better performance and make sampling downconversion attractive for many radar, communications, electronic-warfare (EW), and electronic intelligence (ELINT) applications. In the sampling downconversion process (Fig. 4), the RF signal is sampled by the downconverting sampler and translated to a pulsed output, with amplitude proportional to the amplitude of the input RF signal at the sampling instant. The output or IF impulse is then amplified and sent to an ADC with a built-in T/H amplifier timed to hold the pulse peak.
The firm's advances have led to significantly improved sampling downconverter performance. Earlier this year (at the IEEE Radar 2005 Conference), the firm introduced the first broadband downconversion sampler modules (DCSMs). These featured DC-to-25-GHz bandwidth, 2 GSamples/s sampling rate, 13-GHz IF bandwidth, and better than –60 dBc typical spurious-free dynamic range (SFDR) at 0-dBm input level. The third-order intercept point for that 0-dBm input level is a healthy +32 dBm. The firm currently offers sampling downconversion modules (model 7600) and boards (model 7620) in VME or compact PCI formats. Each board includes a sampling module, the IF chain, and a power converter.
The company has received very positive feedback from initial customers for the new downconverter products. Stephen Krasznay, Multi-Band SAR program manager from NAVAIR, who is working with the company on implementing these new products, states, "the PSPL Downconverter is the missing link. There has long been a desire within the radar community to directly digitize RF at the antenna. The PSPL Downconverter will finally allow this to be accomplished by directly sampling the high frequency RF (i.e., 10 GHz) with 1 GHz bandwidth and coupling directly to COTS digitizers, all while maintaining high performance."
The firm's samplers can also be used to extend the range of a given ADC. ADCs suffer SFDR degradation as the input signals approach the full-scale limits of the converter and the analog bandwidth of the T/H amplifier ahead of the ADC, due to slew-rate limitations and the dynamic nonlinearity of the T/H amplifier. These nonlinearities are frequency dependent and not easily correctable with digital signal processing (DSP). By using one of the company's samplers in front of the T/H amplifier, the ADC is presented with a signal shape independent of input frequency (since the ADC is processing the IF from the sampler). This effectively renders the dynamic nonlinearities of the ADC as static nonlinearities, which can be corrected by DSP (Fig. 5).
Broadband sampling upconversion utilizes a very similar architecture to the sampling downconverters (Fig. 6). In this example single-channel real-time system, a sampling pulse modulator is used with a DSP, digital-to-analog converter (DAC), and IF signal processing. Digital baseband signals are fed to the DAC and modulate the envelope of an alternating phase (monocycle) pulse train. The sampler strobe consists of a NLTL-based pulse-forming network driven from either a fundamental or multiplied LO sinewave. The shape and time offset of the alternating polarity monocycle RF pulses are tailored to maximize the spectral power in the desired output frequency band. The firm has sucessfully built and characterized prototypes of a sampling upconverter and plans to introduce products in the near future.
NLTL circuits not only make good sampling strobe generators, they also make good comb generators. The most commonly used semiconductor device for comb generation has historically been the SRD, although they have limitations, including phase noise. SRD devices create a fast edge and generate harmonics by sweeping stored minority carriers from the depletion region during the transition from forward to reverse bias with large reverse recovery currents. As a result, SRDs are subject to recombination noise as well as shot noise. In the time domain, these processes add timing jitter to the output pulse, which manifests itself in the frequency domain as additional phase noise above the traditional baseline 20logN increase (where N is the multiplication factor) of an ideal frequency multiplier. This can add 6 dBc/Hz or more to the output phase noise.
NLTLs employ a completely different physical mechanism for frequency multiplication than an SRD-based component, more closely related to Schottky varactor based frequency multipliers. Being basically passive (albeit nonlinear) majority carrier devices, NLTLs do not suffer from recombination and shot noise, and consequently have much better phase-noise performance.
In addition to low phase noise, NLTL-based comb generators also offer several other advantages over narrowband SRD-based comb generators. Typically, SRD-based comb generators have less than a five- percent input frequency range and are available only for specific frequencies. Since NLTL lines are inherently broadband, they can accommodate input ranges well over an octave in a single device. In addition, since NLTLs have faster transitions, and can be driven at higher input frequencies than SRD comb generators and they have more power in the higher harmonics. NLTL-based comb generators are useful well into W-band (70 to 100 GHz).
Earlier this year, the company introduced a line of comb generators based on NLTL technology. These comb generators cover an input range of 80 MHz to 2 GHz and provide outputs to 50 GHz. A comparison of the measured residual phase noise for a commercial SRD comb generator and one of the firm's NLTL comb generators is shown in Fig. 7. The measured phase noise of the NLTL for the 10th harmonic is at the thermal noise floor of the phase-noise measurement system (–177 dBc/Hz at 200 MHz). Picosecond Pulse Labs, 2500 55th St., Boulder, CO 80301; (303) 443-1249, FAX: (303) 447-2236, e-mail: kschoen@picosec ond.com, Internet: www.picosecond.com.
- M. Kahrs, "50 years of RF and Microwave Sampling," IEEE Transactions on Microwave Theory and Techniques, Vol. 51, No. 6, June 2003, p.1787-1805.
- Agoston, A.; Pepper, S.; Norton, R.; Ebner, J.; Schoen, K "100 GHz through-line sampler system with sampling rates in excess of 10 Gsamples/second," 2003 IEEE MTT-S International Microwave Symposium Digest, Vol. 3, p. 1519- 1521.
- Pepper, S., "Real-Time Sampling Downconverter Front Ends for Digital Radar and Wide-Band Signaling," 34th European Microwave Conference, 12 to 14 October 2004, Amsterdam, The Netherlands.
- US Patent applications 20050128020, 20020145484, PCT/US/2005/027586.
- Rodwell, Mark J.W. et al., "GaAs nonlinear transmission lines for picosecond pulse generation and millimeter-wave sampling," IEEE Transactions on Microwave Theory and Techniques (ISSN 0018-9480), Vol. 39, July 1991, p. 1194-1204.