A variety of ultrawideband system designs — from carrierless, impulse UWB (I-UWB) to multicarrier (MC) UWB — have been pursued recently. I-UWB has several advantages, including simple RF circuitry, robustness to multipath fading and subcentimeter ranging ability. These advantages benefit such applications as home networking or wireless sensor networks. The wideband nature of I-UWB, however, tests the limits of current technology with three specific challenges: sampling, harvesting multipath energy and detecting the presence of I-UWB signals to provide clear-channel assessment (CCA).

Sampling extremely narrow pulses that occupy several gigahertz of bandwidth requires exceptionally high sampling speed; sampling multiple paths demands complex circuitry; and detecting a narrow pulse within a large search interval is time-consuming. These problems can be addressed through realization of a simple principle in communications: the duality of time and frequency. An I-UWB signal, which is present for only a short time, spreads its power over an ultrawide spectrum. Where traditional time-domain approaches to process such a narrow pulse encounter considerable challenges, a frequency-domain approach provides a solution. It can enable a digital CMOS implementation with moderate circuit complexity and power dissipation while also improving bit error rate, extending link distance and addressing such design challenges as channel estimation and compensation.

A signal may be represented in either the frequency or the time domain, and communications systems normally reconstruct a signal from discrete samples in time. I-UWB systems communicate with a train of narrow pulses that can occupy between 500 MHz and 7.5 GHz of bandwidth, and the theoretical minimum sampling rate is at least twice that bandwidth. Analog-to-digital converters operating between 1 and 15 GHz push the limits of existing CMOS technology and dissipate enormous power. Therefore, existing I-UWB receivers process the received signal with analog correlators in silicon germanium technology.

Frequency-domain sampling supports CMOS A/D converter implementation by relaxing the sampling rate, decreasing circuit complexity and reducing power dissipation. Instead of reconstructing the signal from discrete samples in time, spectral samples (each containing a frequency, a phase and a magnitude) represent each received symbol. The receiver captures harmonic spectral components of a signal that falls within a time window longer than the pulse width and shorter than the pulse repetition interval (PRI). The spectral components captured during the time window describe the received signal as Fourier series coefficients. Since the time window extends beyond the desired received signal, any discontinuous points at the window boundaries can be safely discarded.

The length of the time window determines the fundamental and harmonic frequencies of the Fourier series coefficients, and hence determines the number of A/Ds. For example, with a 1-nanosecond time window for an I-UWB pulse with a 7-GHz bandwidth and 100-ns PRI, the resulting fundamental frequency is the inverse of the time window (1 GHz), and the harmonic frequencies are multiples of the fundamental frequency. Two second-order filters extract the complex valued spectral components. Thus, the receiver requires 14 filters and A/Ds to sample seven real and seven imaginary coefficients at 4 GHz, 5 GHz, . . . 10 GHz.

The sampling rate depends only on the PRI; here, it's 10 Msamples/second (1/PRI). Within a time window, each set of spectral samples represents a portion of the received signal as a continuous waveform. For efficient operation, the receiver may operate on the spectral samples; if necessary, a zero-padded n-point inverse Fourier transform (IFT) reconstructs the signal in the time domain.

Second, the narrow pulses result in many resolvable multipaths. A conventional narrowband receiver harvests and constructively combines multipath energy with a rake receiver. One rake finger processes each path, so the number of fingers fixes the number of harvested multipaths and the performance. But each rake finger increases circuit complexity, power dissipation and storage requirements.

When harvesting multipath energy in the frequency domain, circuit complexity is independent of the number of multipaths within the time window. Spectral samples represent all multipath energy within the time window without tracking and deskewing individual multipaths.

During channel estimation, the multipath energy harvester identifies paths by multiplying the received signal with a sliding template of a pulse in the frequency domain. In the time domain, the multiplication result is a correlation function that characterizes the channel impulse response with the arrival time of all multipath signals that exceed a threshold. The impulse response yields a multipath template, which is converted to the frequency domain.

During operation, the receiver multiplies the spectral samples by the multipath template. This simple operation results in simpler hardware than a correlation operation. Further, a single set of coefficients represents multipath energy, so the frequency-domain energy harvester does not require the additional fingers, samples and processing speed of a rake receiver.

Finally, the narrow pulses and low radiated power of I-UWB present difficulties in detecting a busy medium to provide CCA, which is fundamental to random, distributed media-access protocols. CCA checks for a free channel before transmitting and initiates reception when it senses the start of a packet.

The frequency-domain approach provides a practical CCA service for I-UWB networks to permit distributed, random access. The CCA circuit examines the received spectral energy, which is always present during a transmission, to avoid searching for a narrow pulse in the time domain. The circuit is moderate in hardware complexity and power dissipation as compared with a full receiver.

Frequency-domain CCA detects activity within a short time, since the filters start oscillation within the first few pulses of a transmission and the energy detectors reach their maximum values well within one PRI. Multipath effects have little influence on the probability of detection, since the multipaths are in-band and repeat at the PRI. Timing jitter does not significantly change the location of the spectral lines, so it does not affect performance.

Nathaniel J. August(nate_august@yahoo.com) is a co-author of An Introduction to Ultra Wideband Communication Systems (Prentice-Hall) due out in September. Hyung-Jin Lee (hlee@vt.edu) is a PhD candidate in the VLSI for Telecommunications lab at Virginia Polytechnic Institute's Bradley Department of Electrical and Computer Engineering (Blacksburg, Va.).

See related chart