Method and System for Estimating Time of Arrival of Signals Using Multiple Different Time Scales

Abstract

A method and system (500) estimates a time of arrival of a signal received via a wireless communications channel. The energy in the received signal is conditioned at multiple different time scales to produce a conditioned signal (700). Then, a leading edge is detected in the conditioned signal (550). The leading edge corresponds to the time of arrival of the received signal (551).

Description

BACKGROUND OF THE INVENTION

The present invention relates generally to radio communication systems, and more particularly to determining a time of arrival of a received signal in a wireless communications network for radio ranging applications.

Ranging

To estimate a distance between a transmitter and a receiver in a wireless communications network, the transmitter sends a signal to the receiver at a time instant t₁ according to a clock of the transmitter. After receiving the signal, the receiver immediately returns a reply signal to the transmitter. The transmitter measures a time of arrival (TOA) of the reply signal at a time t₂. An estimate of the distance between the transmitter and the receiver is the time for the signal to make the round trip divided by two and multiplied by the speed of light

$c,i.e.,D=\frac{t_1-t_2}{2}c.$

This is also known as ‘ranging.

Matched Filtering

In a conventional ranging system as shown in FIG. 1A, a signal 101 received at an antenna is pre-filtered 200 and passed to a matched filter 300. Leading edge detection 150 can be performed on the output of the matched filter.

As shown in FIG. 2, a typical pre-filter 200 includes a linear low noise amplifier (LNA) 210, and a band-pass filter (BPF) 220. Generally, the output of the matched filter 300 is sampled at a Nyquist rate to produce discrete signal decision statistics for the edge detection.

As shown in FIG. 3, the matched filter 300 can use a time shifted 301 template signal 310 to produces a maximum correlation with the received signal 110 and by applying an integrator 320. Using a sampler circuit 330, a highest peak at the output of the matched filter 300 is considered the TOA estimate by the signal edge detector 150, G. L. Turin, “An introduction to matched filter,” IRE Trans. on Information Theory, vol. IT-6, no. 3, pp. 311-329, June 1960. The time shifted template signal is adjusted 301 adaptively. In other words, correlations of the received signal with shifted versions of the template signal 310 are considered. In a single path channel, the transmitted waveform can be used as the optimal template signal, and conventional correlation-based estimation can be employed.

However, in the presence of an unknown multipath channel, the optimal template signal becomes the received waveform, which is a convolution of the transmitted waveform with the channel impulse response. Therefore, the correlation of the received signal with the transmit-waveform template is suboptimal in a multipath channel. If this suboptimal technique is employed in a narrowband system, the correlation peak may not give the true TOA because multiple replicas of the transmitted signal partially overlap due to multipath propagation.

In order to prevent this effect, super-resolution time delay estimation techniques have been described, M.-A. Pallas and G. Jourdain, “Active high resolution time delay estimation for large BT signals,” IEEE Transactions on Signal Processing, vol. 39, issue 4, pp. 781-788, April 1991.

Edge Detection on an Over-Sampled Signal

As shown in FIG. 1B, signal processing for edge detection 150 can be performed directly on the output of the pre-filter 120. However, the high sampling rate makes the method impractical for real-time applications.

Signal Processing Prior to Step Detection

As shown in FIG. 1C for other conventional applications 170, e.g., electrocardiograms (ECG), acoustic signals and images, the signals are first processed 400 so that edge detection 190 can distinguish weak first arrival paths from noise by enhancing peaks due to the signal, and suppressing peaks associated with noise. As shown in FIG. 4A, a bank 410 of wavelet filters φ 420 can be applied to the signal to enhance edge detection.

Leading Edge Detection

Detecting leading edges of signals has analogies with various other fields including: object edge detection in image processing, J. Canny, “A computational approach to edge detection,” IEEE Trans. Pattern Anal. Machine Intel, vol. 8, pp. 679-698, 1986, and H. Moon, R. Chellappa, and A. Rosenfeld, “Optimal edge-based shape detection,” IEEE Trans. Image Proc., vol. 11, no. 11, pp. 1209-1227, November 2002; voice activity detection in speech processing, S. G. Tanyer and H. Ozer, “Voice activity detection in non-stationary noise,” IEEE Trans. Speech and Audio Processing, vol. 8, no. 4, pp. 478-482, July 2000, A. Q. Z. Qi Li; Jinsong Zheng; Tsai, “Robust endpoint detection and energy normalization for real-time speech and speaker recognition,” IEEE Trans. Speech and Audio Processing, vol. 10, no. 3, pp. 146-157, March 2002, and J. Sohn, N. S. Kim, and W. Sung, “A statistical model-based voice activity detection,” IEEE Signal Processing Lett., vol. 6, no. 1, pp. 1-3, January 1999; and spike-detection in biomedical engineering, Z. Nenadic and J. W. Burdick, “Spike detection using the continuous wavelet transform,” IEEE Trans. Biomedical Engineering, vol. 52, no. 1, pp. 7487, January 2005, S. Mukhopadhyay, G. C. Ray, “A new interpretation of nonlinear energy operator and its efficacy in spike detection,” IEEE Trans. Biomedical engineering, vol. 45, no. 2, pp. 180-187, February 1998; and electrocardiograms, C. Li, C. Zheng, and C. Tai, “Detection of ECG characteristic points using wavelet transforms,” IEEE Trans. BiomedicalEngineering, vol. 42, no. 1, pp. 21-28, January 1995)

Detecting drastic changes in signals is described extensively in the prior art. When statistics of the signal are known before and after a change-point, an optimal detection can be achieved by tracking log-likelihood ratios of the signals from two hypothesized distributions.

Considering more basic techniques, a simplest approach for detecting edges of a signal is to pass the signal through a gradient operator, such as [−1 0 1]). However, this technique does not consider the effects of noise. The performance of the gradient operator can be improved by using a filtered derivative techniques for smoothing.

Scale-space filtering, as shown in FIG. 4A, was first described by A. Witkin, “Scale-space filtering: A new approach to multi-scale description,” Proc. IEEE Int. Conf. Acoust., Speech, Signal Processing (ICASSP), vol. 9, pp: 150-153, March 1984. The input signal 170 is smoothed at various time scales with Gaussian distributions of different variances φ 420. Local minima and maxima of the derivative of the smoothed signal, at various time scales, which can also be obtained by filtering an initial signal with derivatives of Gaussian distributions at various time scales, correspond to the edges of the signal at different scales. Zero-crossings of a convolution of the signal with the second derivatives of Gaussian distributions, at various scales, can identify the edges. However, this technique does not reveal a direction of the edge, i.e., whether the edge is a rising-edge or a falling-edge, nor, a sharpness of the edge. Witkin describes a coarse-to-fine tracking of the edges in the scale-space image, by exploiting the correlations across the scales, to identify and localize major singularities in the signal.

The scale-space representation of signals uses a wavelet-theory framework, and a wavelet transform modulus maxima (WTMM) for the identification of major edges in the signal is used by A. Mallat and W. L. Hwang, “Singularity detection and processing with wavelets,” IEEE Trans. Information Theory, vol. 38, no. 2, pp. 617-643, March 1992, and A. Mallat and S. Zhong, “Characterization of signals from multi-scale edges,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 14, no. 2, pp. 710-732, July 1992.

By analyzing an evolution of the wavelet transform exponent across scales, local Lipschitz exponent, which measure a local regularity of the signal, can be estimated. That effectively ‘denoises’ the signal using the Lipschitz exponent, and other a priori information.

A direct multiplication of wavelet transform data, at various scales, can be used to enhance signal edges and suppress the noise, Y. Xu, J. B. Weaver, D. M. Healy, and J. Lu, “Wavelet transform domain filters: a spatially selective noise filtration technique,” IEEE Trans. Image Processing, vol. 3, no. 6, pp. 747-758, July 1994.

Using the product of multi-scale wavelet coefficients to detect sharp edges in signals is described by A. Swami and B. M. Sadler, “Steps change localization in additive and multiplicative noise via multi-scale products,” Proc. IEEE Asilomar Conf. Signals, Systems, Computers, vol. 1, pp. 737-741, November 1998, B. M. Sadler and A. Swami, “On multi-scale wavelet analysis for step estimation,” Proc. IEEE Int. Conf. Acoust., Speech, Signal Processing (ICASSP), vol. 3, pp. 1517-1520, May 1998, S. MacDougall, A. K. Nandi, and R. Chapman, “Multiresolution and hybrid Bayesian algorithms for automatic detection of change points,” IEEE Proceedings-Vision, Image, and Signal Processing, vol. 145, no. 4, pp. 280-286, August 1998, J. Ge and G. Mirchandani, “Softening the multi-scale product method for adaptive noise reduction,” Proc. IEEE Asilomar Conf. Signals, Systems, Computers, vol. 2, pp. 2124-2128, November 2003, L. Zhang and P. Bao, “A wavelet-based edge detection method by scale multiplication,” Proc. IEEE Int. Conf. Pattern Recognition, vol. 3, pp. 501-504, August 2002, and M. Beauchemin and K. B. Fung, “Investigation of multi-scale product for change detection in difference images,” Proc. IEEE Int. Geoscience and Remote Sensing Symp. (IGARSS), vol. 6, pp. 3853-3856, September 2004.

Ultra Wideband

Ultra wideband signals are drastically different than conventional wireless signals. Not only is the signal spread over a huge frequency range, but in addition, the extremely short pulses that constitute the signal are also spread out over time. For example, the signal can cover anywhere from 500 MHz to several GHz of the radio spectrum, and bursts of ultra-low power pulses are often in the picosecond, i.e., 1/1000th of a nanosecond, range. The pulses are transmitted across all frequencies at once. Furthermore, UWB signals are subject to dense multipath propagations.

FIG. 4B compares the potential effect of the differences between a conventional signal 451 and an ultrawideband signal 452. Missing the peak at time 461 for the conventional signal 451 will have minimum effect on estimating the correct TOA, while missing the peak at time 462 for the UWB signal 452 can be completely erroneous.

However, as the bandwidth of the UWB signal increases, the signal is less spread in time and a rising edge of the received signal becomes sharper. In precision ranging applications, detecting the arrival time of the rising edge of the received signal at desired accuracies is important. Therefore, it is desired to use UWB signals to provide precise positioning capabilities.

Prior art matched filters are described by W. Chung and D. Ha, “An accurate ultra wideband (UWB) ranging for precision asset location,” Proc. IEEE Conf. Ultrawideband Syst. Technol. (UWBST), pp. 389-393, November 2003, B. Denis, J. Keignart, and N. Daniele, “Impact of NLOS propagation upon ranging precision in UWB systems,” Proc. IEEE Conf. Ultrawideband Syst. Technol. (UWBST), pp. 379-383, November 2003, and K. Yu and I. Oppermann, “Performance of UWB position estimation based on time-of-arrival measurements,” Proc. IEEE Conf. Ultrawideband Syst. Technol. (UWBST), pp. 400-404, May 2004.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1A is a block diagram of prior art leading edge detector with a matched filter;

FIG. 1B is a block diagram of a prior art leading edge detector without a matched filter;

FIG. 1C is a block diagram of a prior art leading edge detector with a signal conditioner;

FIG. 2 is a block diagram of a pre-filter for the leading edge detector of FIG. 1A;

FIG. 3 is a block diagram of a matched filter for the leading edge detector of FIG. 1A;

FIG. 4A is a block diagram of the signal conditioner of the leading edge detector of FIG. 1C;

FIG. 4B compares peaks in conventional signals with peaks in ultra wideband signals;

FIG. 5A is block diagram of a ranging system and method;

FIG. 5B is a timing diagram of a transmitted time hopping impulse radio signal to be detected;

FIG. 6A is a block diagram of the signal energy collector that produces observation samples using signal energy collection;

FIG. 6B is a block diagram of the signal energy collector that produces observation samples using matched filtering;

FIG. 6C is a block diagram of the signal energy collector that produces observation samples using transmitted reference structure;

FIG. 7 is a detailed block diagram of a ranging method using a wavelet filter bank; and

FIG. 8 is a detailed block diagram of a ranging method using a multi-scale filter bank.

System Structure and Method Operation

As shown in FIG. 5A, we provide a system and method 500 for estimating a time of arrival (TOA) of a signal 501 received via a wireless channel 502 at a radio transceiver in a wireless communications network. The TOA in a ranging application can be used to determine a distance between two transceivers. For the purpose of this description, the transceiver is estimating the TOA for a received signal. However, it should be understood that the transceiver can transmit and receive. In a preferred embodiment, the received signal 501 is an ultra wideband radio signal.

The signal 501 is received at an antenna of a transceiver. Energy of the signal is collected 600. The collected signal energy is conditioned 700 using multiple different time scales. Furthermore, the conditioning can be improved by considering characteristics 520 of the channel and signal parameters 540. Leading edge detection 550 is then preformed on the condition signal energy to estimate the TOA 551.

Our method is significantly different than the prior art. We use multiple different time scales. In one embodiment, we apply products of multi-scale wavelet coefficients during the signal conditioning 700, see FIG. 7. In an alternative embodiment, we apply multi-scale filters during the conditioning, see FIG. 8. Multiple different time scales have never been used for conditioning a collected energy of a radio signal to perform leading edge detection for the purpose of radio ranging.

The signal energy conditioner 700 takes the channel characteristics 520, e.g., signal to noise ratio (SNR), etc., and the signal related parameters 540, e.g., bandwidth, frame, symbol, chip, and block lengths the signal, into consideration to improve a performance of the energy edge detector 550. The signal related parameters 540 can be used to select appropriate wavelet time scaling filters and coefficients of the multiple different time scales, while the channel characteristics 520 are used in selecting the different time scales involved in the product calculation.

At relatively low SNRs, e.g., less than 20 dB, fine time scales are excluded from the product determinations, because a high noise level cannot be smoothed out at fine time scales. Therefore, the fine time scales, unless they are removed, can cause erroneous peaks.

We use multi-time scale analysis of the received signal energy to conditioning the signal. The conditioning enhances peaks relatively near to the leading edge of the received signal 501, and suppresses noise. Then, edge detection methods can be applied to the output of the conditioner 700.

Signal Model

The received signal can be an ultra-wideband signal (UWB). Specifically, a time-hopped impulse-radio signal (TH-IR). However, it should be understood, that the signal can be of other forms, as known in the art.

As shown in FIGS. 5B-5C, for the ultra wideband signal 501, wireless impulse radio transceivers allocate time in terms of symbol time (T_S) 595, frame times (T_F) 575, blocks times (T_B) 585, and chip times (T_C) 565. Frames are longer than blocks, which are longer than chips. Each frame can include multiple blocks. Each block can include multiple chips, and each symbol (T_S) 595 can include multiple frames 565.

As shown in FIG. 5B, a single radio pulse 555 is transmitted in each frame 565 within a block 585 at a predetermined position (time) in a chip 565. As shown in FIG. 5C, multiple pulses 555 and 556 are transmitted for each symbol. The later pulses are delayed (D) 640 versions of the first pulse. It should be understood that more than two pulses can be transmitted for each symbol.

The predetermined position of the single pulse or multiple pulses can be different for different symbols. Typically, the position of the pulse in the frame indicates the value of the symbol. The received time hopping (TH) impulse radio (IR) signal 501 can be represented by

$\begin{array}{cc}r\left(t\right)=\sum _{j=-\infty }^{\infty }d_j{\omega }_{mp}\left(t-{\mathrm{jT}}_f-c_jT_c-{\tau }_{\mathrm{toa}}\right)+n\left(t\right),& \left(1\right)\end{array}$

where a frame index is j, a frame duration is T_f, a number of pulses per symbol is N_s, a chip duration is T_c, a symbol duration is T_s, the TOA of the received signal is τ_toa, and a possible number of chip positions per frame N_h is given by N_h=T_f/T_c. An effective pulse after the channel impulse response is given by

$\begin{array}{cc}{\omega }_{mp}\left(t\right)=\sqrt{E}\sum _{l=1}^L{\alpha }_l\omega \left(t-{\tau }_l\right),& \left(2\right)\end{array}$

where the received UWB pulse is ω(t) is a pulse energy E, a fading coefficients α_i, and delays of the multipath components are τ₁. Additive white Gaussian noise (AWGN) with zero-mean and double-sided power spectral density N₀/2 and variance σ² is denoted by n(t). No modulation is considered for the ranging process.

In order to avoid catastrophic collisions, and smooth the power spectral density of the transmitted signal, time-hopping codes c_j^(k), that can take values in {0, 1, . . . , N_h−1} are assigned to different transceivers. Moreover, random-polarity codes d_j{=±1} provide additional processing gain for detecting the signal, and smoothing the signal spectrum.

For simplicity of this description, we assume that the signal arrives in one frame duration, i.e., τ_TOA<T_f, and there is no inter-frame interference (IFI), i.e., T_f≧(L+c_max)T_c, or, equivalently, N_c≧L+c_max, where c_max is a maximum value of the TH sequence. It should be understood that multiple frames can be used for a symbol.

Note that the assumption of τ_TOA<T_f does not restrict us. In fact, it is enough to have τ_TOA<T_s to work when the frame is large enough and a predetermined TH codes are used. Moreover, even if τ_TOA>T_s, an initial energy detection can be used to determine the arrival time within a symbol uncertainty.

Signal Energy Collection

UWB signals are quite unlike conventional signals in a number of ways. First the signal is spread over an extremely large frequency range. Second, the pulses that constitute the signal are spread over time. Third, the signal suffers from multipath propagation, and fourth, the amount of energy in the pulses is very low. For example, FCC regulations require UVWB systems to emit energy at less than −41.3 dBm/Hz, over a spectrum from 3.1 GHz to 10.6 GHz. The low power requirement results in increased sensitivity of UWB signals to interference and fading. Therefore, in the case the signal is a wideband signal, we provide three different ways that energy can be collected 600 in an optional step to produce the signal for the energy conditioner 700.

Square-Law Device

As shown in FIG. 6A, the energy collector 600 includes a linear low noise amplifier (LNA) 601, a band-pass filter (BPF) 605, a square-law device 610, an integrator 620, and a sampler circuit 630, serially connected to each other. The square-law device outputs a squared form of an input signal, as known in the art. The sampling circuit uses a sampling interval of t_s, which is equal to the block length T_B, The output of the integrator 620 are observation samples z(n), as analytically expressed as:

$\begin{array}{cc}z\left[n\right]=\sum _{j=1}^{N_s}{\int }_{\left(j-1\right)T_f+\left(c_j+n-1\right)T_b}^{\left(j-1\right)T_f+\left(c_j+n\right)T_b}{r\left(t\right)}^2t& \left(3\right)\end{array}$

Signal parameters 540 can include signal bandwidth, frame duration 575, block duration 585, chip duration 565, and symbol duration 595, etc. The integrator 620 intervals are determined according to the signal parameters 520.

Stored-Reference

As shown in FIG. 6B, for the collector 600′, with a sampling interval of t_s, which is equal to block length T_b, the received signal 501 is correlated 615 with a stored reference signal 503. The stored reference signal can be pre-stored in the receiver. The resulting output is sampled at by sampler circuitry 630 according to

$\begin{array}{cc}{s}_{\mathrm{tmp}}\left(t\right)=\sum _{j=0}^{N_s-1}d_j\omega \left(t-{\mathrm{jT}}_f-c_jT_c\right),& \left(4\right)\\ z_n^{\left(\mathrm{sr}\right)}={\int }_{\left(n-1\right)t_s}^{\left(n-1\right)t_s+N_sT_f}r\left(t\right)s_{\mathrm{tmp}}\left(t-\left(n-1\right)t_s\right)t,& \left(5\right)\end{array}$

Transmitted Reference

As shown in FIG. 6C, for the collector 600″, with a sampling interval of t_s, which is equal to the block length T_B, the received signal is correlated 616 with a delayed version (D) 640 of the signal, see FIG. 5C. The resulting output is sampled at the sampler circuitry 630 according to

$\begin{array}{cc}{z}_n^{\left(\mathrm{tr}\right)}=\sum _{j=1}^{N_s}{\int }_{\left(j-1\right)T_f+\left(c_j+n-1\right)t_s}^{\left(j-1\right)T_f+\left(c_j+n\right)t_s}\tilde{r}\left(t\right)\tilde{r}\left(t-D\right)t,& \left(6\right)\end{array}$

Signal Conditioner

As shown in FIG. 7, the signal conditioner 700 enhances the signal and suppresses noise prior to energy edge detection 550. The signal parameters 540 can be used to optimize selection of wavelet filter types 720 and coefficients for the time scaling in the wavelet filter bank 710. Wavelets are a class of function used to localize a given function in both space and scaling. Channel characteristics 520, such as an estimated SNR 521 are provided as a feedback to a branch selector 740 for selecting the appropriate wavelet filter bank outputs for a product determination 745. At a low SNR, erroneous peaks can occur. This is due to the fact that the fine scale wavelet filters maintain high frequency components of a high noise level. These peaks can distort the output of the product 745, and can be misleading in the signal energy edge detection 550. Therefore, if the SNR level is low, then the fine time scales are removed from the product determination in the signal energy conditioner 700. The smoothing branch 730 is used for noise suppression, and it typically contains a filter function in the form of a Gaussian curve.

The product output is fed to the signal energy edge detector 550. The signal energy edge detector 550 can use conventional edge detection techniques, as known in the art, such as threshold-based, threshold based with search back, maximum likelihood based, and the like.

Alternatively, signal energies from coarse to fine time scales in a multi-scale filter bank 810 can be used to improve the performance of the leading edge detector 550. In FIG. 8, the filters are arranged, from the bottom to the top, in a coarse to fine time scale order. The coarser filters have a greater smoothing effect than the finer filters. For example, the coarsest filter can consider 64 samples over a relatively large amount time, and the finest filter considers only one sample over a small amount of time. Of course, the exact amount of time, for a particular application, depends on the signaling frequency. If the SNR is less than 20 db, then the finest time scale is excluded.

Because the energy values at different time scales are correlated, their product is expected to enhance the peaks due to signal existence. A rectangular filter h₂·[n] at a time scale s is given by

h₂·[n]=u[n+2^p]=u[n],

where s=1, 2, . . . , S is a time scale number ranging from finer scales to coarser scales, S=└log₂Nb┘, and u[n] is a step function. A convolution of h₂·[n] with the energy vector z produces energy concentration of our signal at various time scales, given by

$y_s\left[n\right]=\sum _kz\left[k\right]h_2·\left[n-k\right].$

Because the values y_s[n] are correlated across multiple different time scales, we can use direct multiplication 815 to enhance the peaks closer to the leading edge of the signal, and suppress noise components, i.e.,

$P_S^{\left(\mathrm{MEP}\right)}\left[n\right]=\prod _{s=1}^Sy_s\left[n\right],$

where P_S^(MEP)[n] denotes the product of convolution outputs from scale l, which is the energy vector itself, through scale S, which is the output of the conditioner 800. Then, the location (time) of the strongest path is estimated as

${\hat{t}}_{\mathrm{MEP}}=\left[\underset{l\le n\le N_s}{\arg \max }\left\{P_S\left[n\right]\right\}\right]T_b$

by a global maxima detector 820.

After the strongest energy block is estimated, a search-back process in the detector 550 can accurately estimate the time of arrival of the leading edge of the signal. Therefore, a search back window length 840 is provided to the leading edge detector 550.

It is to be understood that various other adaptations and modifications may be made within the spirit and scope of the invention. Therefore, it is the object of the appended claims to cover all such variations and modifications as come within the true spirit and scope of the invention.

Claims

1. A method for estimating a time of arrival of a radio signal, comprising:

conditioning, at multiple different time scales, an energy of a received signal to produce a conditioned signal; and

detecting a leading edge of the conditioned signal, the leading edge corresponding to a time of arrival of the received signal.

2. The method of claim 1, further comprising:

estimating a distance between two transceivers based on the time of arrival.

3. The method of claim 1, in which the received signal is an ultra wideband radio signal.

4. The method of claim 1, in which the received signal is a time-hopped, impulse-radio signal.

5. The method of claim 1, in which the signal is received via a wireless channel, and further comprising:

conditioning the received signal based on characteristics of the wireless channel.

6. The method of claim 1, further comprising:

conditioning the received signal based on parameters of the signal.

7. The method of claim 1, in which the signal is received via a wireless channel, and further comprising:

conditioning the received signal based on characteristics of the wireless channel and parameters of the signal.

8. The method of claim 1, further comprising:

applying multi-scale wavelet coefficients to the energy of the signal during the conditioning.

9. The method of claim 1, further comprising:

applying multi-scale filters to the energy of the signal during the conditioning.

10. The method of claim 1, further comprising:

receiving the signal; and

collecting the energy of the received signal.

11. The method of claim 5, in which the characteristics of the channel include a signal to noise ratio.

12. The method of claim 6, in which the parameters of the received signal include a bandwidth, a frame length, a symbol length, a chip length, and a block length of the received signal.

13. The method of claim 6, further comprising:

selecting wavelet scaling filters and coefficients of the multiple different time scales according to the parameters of the received signal.

14. The method of claim 1, further comprising:

excluding fine time scales during the conditioning at low signal to noise ratios.

15. The method of claim 1, in which the conditioning enhances peaks relatively near the leading edge.

16. The method of claim 4, in which multiple pulses are transmitted for each symbol in the received signal.

17. The method of claim 16, in which the multiple pulses include delayed versions of a first one of the multiple pulses.

18. The method of claim 1, in which the time of arrival is less than a length of a symbol represented by the received signal.

19. The method of claim 10, in which collecting uses a square-law device.

20. The method of claim 10, in which collecting uses a stored reference signal.

21. The method of claim 10, in which collecting uses a transmitted reference signal.

22. A system for estimating a time of arrival of a radio signal, comprising:

means for conditioning, at multiple different time scales, an energy of a received signal to produce a conditioned signal; and

means for detecting a leading edge of the conditioned signal, the leading edge corresponding to a time of arrival of the received signal.