MULTIPLE DESCRIPTION CODING OVER MULTIPLE TRANSMISSION RESOURCES IN TIME OR FREQUENCY USING ANALOG MODULATION Field of the Invention The present invention relates generally to the field of signal processing, and more particularly relates to multiple description coding of signals for transmission over a communication network or other type of communication medium. Background of the Invention In a typical multiple description coding arrangement, a given signal to be transmitted is processed in a transmitter to generate multiple descriptions of that signal, and the multiple descriptions are then transmitted over a network or other communication medium to a receiver. Each of the multiple descriptions may be viewed as corresponding to a different transmission channel subject to a different loss probability. The goal of multiple description coding is generally to provide a signal reconstruction quality at the receiver that improves as the number of received descriptions increases, without introducing excessive redundancy between the various multiple descriptions. One known multiple description coding technique is commonly referred to as quantized frame expansion. The signal to be transmitted may be represented as an N-dimensional symbol vector x = { i, x2, ..., X } The symbol vector x is multiplied by a frame expansion transform T to generate an M-dimensional symbol vector^ = Tx = {yi, y 2, ..., ¾ }, where the transform J is an M xN matrix and M >N . The symbol vector^ is then subj ect to a quantization operation to form 7 = Q(y). Forward error correction (FEC) and cyclic redundancy check (CRC) codes are then applied to 7 before it is transmitted over a network to the receiver. At the receiver, the received signal 7 is subject to FEC decoding and the CRC is used to detect symbol errors. The symbols with no errors are used to reconstruct an estimate of x . For additional details regarding this and other conventional multiple description coding techniques, see Vivek K Goyal, "Multiple Description Coding: Compression Meets the Network," IEEE Signal Processing Magazine, September 2001, pp. 74-93. Conventional multiple description coding techniques generally assume that the channels are so-called "erasure" channels. With such channels, a given symbol or other piece of data is known to the receiver to be either correct or in error, and some mechanism is needed to provide this capability, such as the above-noted FEC or CRC codes. However, the FEC or CRC codes are useful only for error detection and correction, and cannot otherwise be used to enhance the quality of a reconstructed signal when no errors occur. Use of such codes therefore represents a waste of bandwidth in any channels that do not have errors. U.S. Patent Application Serial No. 12/652,390, filed January 5, 2010 and entitled "Orthogonal Multiple Description Coding," discloses improved multiple description coding techniques that overcome the above-described drawbacks of conventional multiple description coding. In one such technique, multiple descriptions of a given signal are generated by processing the signal using respective ones of a plurality of orthogonal matrices. Each of the multiple descriptions is generated as a function of the signal and a corresponding one of the plurality of orthogonal matrices. For example, M descriptions y ) of an N-dimensional symbol vector x may be generated by applying respective ones of the orthogonal matrices to the vector x in accordance with the following equation: y > =U )x,i = l,...,M where U(i) , i =\,2,...,M denote orthogonal matrices of dimension NxN . The orthogonal matrices introduce redundancy in such a way that the redundancy can be used not only to improve signal reconstruction quality, but also to detect and correct errors in the received signal. The multiple descriptions therefore have error detection and correction capability built into them. This avoids the need to dedicate additional bandwidth for FEC and CRC, thereby ensuring that there will be no wasted bandwidth in the absence of errors, while also providing graceful degradation in the presence of errors. Despite the considerable advantages provided by the above-described orthogonal multiple description coding technique, a need remains for further improvements, particularly with regard to providing optimal coding in the presence of variable channel conditions. For example, in coding techniques in which multiple description coefficients are subj ect to quantization prior to transmission, the bit rate and signal quality is fixed by the quantization level regardless of the actual channel condition. As a result, the bit rate and signal quality may be too low for a good channel, and may be too high for a poor channel. Therefore, such transmissions can lead to either a waste of bandwidth for good channels, or a failure to receive the signal in poor channels. Furthermore, in some systems, the number of transmission subcarriers is required to match the number of coefficients to be transmitted, which unduly limits the applications in which such systems can be used. Summary of the Invention Illustrative embodiments of the present invention provide further improvements in multiple description coding of video and other signals by providing a technique referred to herein as arbitrary precision multiple description coding. In one or more of these illustrative embodiments, the arbitrary precision multiple description coding ensures that bandwidth utilization is optimal across a variety of channel conditions. Thus, in a good channel with higher available bandwidth, bit rate and signal quality are automatically increased. Similarly, in a poor channel with lower available bandwidth, bit rate and signal quality are automatically decreased. The multiple description coding therefore remains optimally matched to the current channel conditions. In accordance with one aspect of the invention, an encoder comprises arbitrary precision multiple description generation circuitry configured to produce multiple descriptions of a given signal by processing the signal using at least one matrix having a dimension which is selected as a function of a designated number of transmission resources, such as orthogonal frequency division multiplexed (OFDM) subcarriers or time division multiplexed (TDM) time slots, that are allocated for transmission of the multiple descriptions. For example, the signal may comprise a vector x of dimension N and the arbitrary precision multiple description generation circuitry may be configured to generate M descriptions of the vector x where the value of M is selected to satisfy a particular one of three possible casesM =N , M >N and M N and M M/2 . It should be noted that the system may include more than M/2 subcarriers although M/2 is assumed to be the number of subcarriers that are currently allocated for transmission of the N coefficients. As indicated previously, other types of modulation, including non-OFDM modulation, may be used in other embodiments. For example, an alternative embodiment utilizing TDM modulation will be described below in conjunction with FIG. 4 . Again, there is no quantization of the entries of the vector s during the modulation process. Instead, the entries of the vector s may be modulated, for example, in floating point format or as very high precision integers. Also, as previously indicated, no additional channel coding is applied. With continued reference to FIG. 3A, after an inverse fast Fourier transform (IFFT) operation is performed on the M/2 subcarriers, the corresponding parallel outputs are then converted to serial form in parallel-to-serial converter 302. The resulting serial stream is further processed in transmission module 304 which performs operations such as, for example, intermediate frequency (IF) conversion, analog-to-digital conversion (ADC), and radio frequency (RF) conversion, and finally transmits the resulting OFDM signal over the wireless network 210. These operations performed in transmission module 304 are exemplary, and additional or alternative operations can be performed in other embodiments. Referring again to FIG. 2, the modulated signal transmitted over wireless network 210 is received in the demodulation module 212, which generates an estimate >of theM-dimensional multiple description vector^: This estimate y is applied to the coefficient reconstruction module 214 which generates an estimate x of the N-dimensional original message x : The original message estimate x is processed in the video reconstruction module to recover the original video signal. FIG. 3B illustrates the operation of the demodulation module 212 in generating the estimate y from an OFDM signal received over the wireless network 210. The received OFDM signal is subject to operations such as RF processing, digital-to-analog conversion (DAC) and baseband processing in reception module 310, and the resulting serial output is converted to parallel form in serial-to-parallel converter 312. An FFT operation is performed to recover the subcarriers that are further processed in module 314 to generate an estimate s of the complex vector s as follows: where each entry of s comprises the complex number associated with a corresponding one of the received subcarriers. The real and imaginary parts of each complex number in s are used to form two real number entries of the estimate y . The estimate y is then utilized along with matrix A =BD to reconstruct the estimate x . FIG. 4A shows the processing of the vector y in the modulation module 206 for transmission over the wireless network 210 in a TDM implementation. The M-dimensional vector^ is initially mapped to a complex vector s of dimension M j 2 , where the complex vector s is again a vector of complex numbers with each such complex number formed from a corresponding pair of real numbers from y . There is no conventional mapping of bits to modulation constellation points in this arrangement. Instead, the real and imaginary parts of the entries of the complex vector s are subject to complex multiplication with respective sin and cos signals in module 400, followed by pulse shaping in module 402. The resulting output is further processed in transmission module 404, which performs operations such as, for example, ADC and R conversion, and finally transmits the resulting TDM signal over the wireless network 210. These operations performed in transmission module 404 are exemplary, and additional or alternative operations can be performed in other embodiments. FIG. 4B illustrates the operation of the demodulation module 212 in generating the estimate y from a TDM signal received over the wireless network 210. The received TDM signal is subj ect to operations such asRF processing and DAC in reception module 410, and the resulting output is subject to further processing in baseband processing module 412. This baseband processing may involve operations such as matched filtering, timing recovery and carrier recovery. The baseband processing module 412 generates at its output the estimate s of the complex vector s, where each entry of comprises the complex number associated with a corresponding one of the received time slots. As in the FIG. 3 embodiment, the real and imaginary parts of each complex number in are used to form two real number entries of the estimate , and the estimate > is then utilized along with matrix A =BD to reconstruct the estimate x . In both the OFDM and TDM implementations of system 200 of FIG. 2 described above, three different cases of relative values of M and N are permitted in the transmitter, namely, M = , M >N and M N and M N , there are more subcarriers or time slots allocated than are needed to transmit the N original message coefficients, and so the N original message coefficients are transmitted with a higher precision. For case M N and M 2 y y +e =Bx +e, x =B y = x +B e . Thus, B is preferably configured such that entries ofB e will be of similar size even if entries of e have different sizes. In the case = N , where the number of allocated subcarriers or time slots exactly matches the number of subcarriers or time slots needed to transmit the N original message coefficients, the matrix B may be an orthogonal matrix of random entries. For example, B may be given by: where v are vectors of length N with random entries. In the case >N , where there are more subcarriers or time slots allocated than are needed to transmit the N original message coefficients, the coefficients can be transmitted with higher precision and the matrix B may be given by: ( ) B = , ( (M) )T where v n are vectors of lengthMwith random entries, and =I, . ., , are orthonormal vectors from v n . As a more particular example for this case, if = kN the matrix B may be given by: B = , U k where U l are orthogonal N N matrices of random entries. This is a type of orthogonal multiple description code as described in the above-cited U.S. Patent Application Serial No. 12/652,390. In the case < N , where there are fewer subcarriers or time slots allocated than are needed to transmit the N original message coefficients, the coefficients can be transmitted with lower precision and the matrix B may be given by: B (u M ) where v are vectors of length N with random entries, and n =I, . ., , are orthonormal vectors from v n . The manner in which the estimate x of the original message vector x is reconstructed from y in coefficient reconstruction module 214 using the matrix^ will now be described in greater detail. This coefficient reconstruction process generally involves finding a solution to the equation: y =Ax , which may be done by finding a least-squares solution in the appropriate one of the three cases M = , M >N and M N , the estimate x may be determined as follows: x = (ATA ATy . Finally, for the case < N , the estimate x may be determined as follows: x = AT (AA T y . It is to be appreciated that other types of coefficient reconstruction techniques may be used in other embodiments. The precision of the reconstructed coefficients can be determined in the following manner. Let x „be the nth coefficient of the original message vector x and let the corresponding reconstructed coefficient of the estimate x be: n = X n + n where e„ is the error in the reconstructed coefficient. Define the precision of the original coefficient xn as: where E(·) denotes the expectation operator. The number of bits of the precision of x „ can then be defined as: \og2Px = -log 2 . Also define the signal-to-noise ratio as E and the bandwidth as B =MIN The theoretical bound for the precision of the reconstructed coefficients is then given The precision of the received coefficients is therefore a monotonically increasing function ofE lN . The corresponding video quality at the output of the video reconstruction module 216 is therefore also a function of E IN as well as the relative values ofMand N. FIG. 5 provides a plot of video quality as a function of signal-to-noise ratio showing performance achievable in an illustrative embodiment at different precision levels, that is, at different relative values ofMand Nas reflected in the bandwidth values B =1, B >1 and Bw <1 which correspond to the respective cases M=N,M>N and MN and M N and M N and M N and M