A system which embeds watermarks in n-dimensional Gaussian data and distributes them in compressed form is studied. The watermarked/compressed data have to satisfy a distortion constraint, and the watermark has to be recoverable in a private scenario (in which the original data are available at the watermark detector). The performance of the system in the presence of additive Gaussian attacks is considered, and the region of achievable quantization and watermarking rate pairs (R/sub Q/,R/sub W/) is established. Moreover, two surprising facts are demonstrated: (1) at low R/sub Q/, the maximum achievable R/sub W/ is the same as when there are no attacks; and (2) at high (but finite) R/sub Q/, the maximum achievable R/sub W/ is the same as when there is no compression (R/sub Q/=/spl infin/). Finally, the performance of related schemes is also discussed.

Two distributed systems are considered for discriminating between two finite-alphabet bivariate memoryless sources and for detecting a known signal in stationary bivariate additive Gaussian noise. Each system comprises two sensors, M-ary local quantizers and a fusion center which makes decisions based on quantized source observations. The problem of asymptotically optimal quantization is considered in detail for the binary (M=2) case. It is shown that optimality is achieved by quantizing a locally computed likelihood ratio wherein one distribution is in general different from the appropriate source marginal. For the problem of detection in Gaussian noise, it is further demonstrated that the optimal distributed system attains the same asymptotic performance as the optimal centralized system for appropriate choice of M.

The Neyman-Pearson performance of a distributed detection system is considered wherein n independent and identically distributed observations are quantized locally into M-ary messages and transmitted to a fusion center. Under fairly general assumptions, it is shown that the type II error probability achieved by the best identical-quantizer system is at most a fixed (in n) multiple of that achieved by the absolutely optimal system.>

Three estimators of the exponent /spl alpha/ in the power spectral density g(/spl lambda/)=c/sub g/|/spl lambda/|/sup -/spl alpha// of fractional Brownian motion are evaluated. These are (i) the periodogram-based estimator /spl alpha//spl circ//sub PG/ (ii) the maximum likelihood estimator /spl alpha//spl circ//sub ML/; and (iii) the Allan (1966) variance-based estimator /spl alpha//spl circ//sub AV/. Large-sample properties of the mean-square error (MSE) and the associated sampling distribution are examined, /spl alpha//spl circ//sub PG/ emphasis on the case /spl alpha//spl isin/(1, 2). The MSE performance of /spl alpha//spl circ//sub PG/ is judged to be inferior to that of both /spl alpha//spl circ//sub ML/ and /spl alpha//spl circ//sub AV/. The rate of decrease of MSE is the same for /spl alpha//spl circ//sub ML/ and /spl alpha//spl circ//sub AV/; the former estimator has smaller MSE, while the latter is less sensitive to departures from the power-law model and is considerably easier to compute.>

The paper considers a binary hypothesis testing system in which two sensors simultaneously observe a discrete-time finite-valued stationary ergodic Markov source and transmit M-ary messages to a Neyman-Pearson central detector. The size M of the message alphabet increases at most subexponentially with the number of observations. The asymptotic behavior of the type II error rate is investigated as the number of observations increases to infinity, and the associated error exponent is obtained under mild assumptions on the source distributions. This exponent is independent of the test level /spl epsiv/ and the actual codebook sizes M, is achieved by a universally optimal sequence of acceptance regions, and is characterized by an infimum of informational divergence rate over a class of infinite-dimensional distributions. Important differences-due to the observations being Markov-between the asymptotically optimal distributed tests and their nondistributed counterparts are highlighted. The converse results require a blowing-up lemma for stationary ergodic Markov sources, which is also proven.>

The performance of a parallel distributed detection system is investigated as the number of sensors tends to infinity. It is assumed that the i.i.d. sensor data are quantized locally into m-ary messages and transmitted to the fusion center for binary hypothesis testing. The boundedness of the second moment of the postquantization log-likelihood ratio is examined in relation to the asymptotic error exponent. It is found that, when that second moment is unbounded, the Neyman-Pearson error exponent can become a function of the test level, whereas the Bayes error exponent remains, as previously conjectured by J.N. Tsitsiklis, (1986), unaffected. Large deviations techniques are also used to show that in Bayes testing the equivalence of absolutely optimal and best identical-quantizer systems is not limited to error exponents, but extends to the actual Bayes error probabilities up to a multiplicative constant.>

The effect of feedback on the performance of a distributed Neyman-Pearson detection system consisting of n sensors, two-stage local quantizers, and a global detector is investigated. Two feedback schemes are discussed, only one of which yields an asymptotic gain in performance, as measured by an appropriate error exponent.>

The asymptotic performance of a multiterminal detection system comprising a central detector and two remote sensors that have access to discrete, spatially dependent, and temporally memoryless observations is discussed. It is assumed that prior to transmitting information to the central detector, each sensor compresses its observations at a rate which approaches zero as the sample size tends to infinity, and that, on the basis of the compressed data from both sensors, the central detector seeks to determine whether the true distribution of the observations belongs to a null class II or an alternative class Xi . Under the criterion that stipulates minimization of the type-II error rate subject to an upper bound epsilon on the type-I error rate, it is shown that, in the case of simple distribution classes ( mod Pi mod = mod Xi mod =1), the error exponent achievable by such a system has a simple characterization, is independent of the value of epsilon , and is insensitive to changes in compression rate as long as the asymptotic rate on one of the sensors is zero.>