WO1998039883A1

WO1998039883A1 - Signalling method using multiple modulus conversion and shell mapping

Info

Publication number: WO1998039883A1
Application number: PCT/US1998/004087
Authority: WO
Inventors: Richard G. C. Williams; Andy L. Norell; Scott A. Lery
Original assignee: 3Com Corporation
Priority date: 1997-03-03
Filing date: 1998-03-03
Publication date: 1998-09-11
Also published as: AU6345198A

Abstract

A frame mapping technique that uses aspect of Multiple Modules Conversion (MMC) and Shell Mapping (SM) to map data bits to a sequence of data symbols, or points. The apparatus includes a modulus converter acting as a Mixed Base Ring Selector (MBRS) to generate ring indices from K data bits. The MBRS is combined with a signal point selector to generate a sequence of transmitted symbols. The signal point selector may operate on modulus conversion principles, trellis coding principles, or uncoded bit mapping techniques. The apparatus and method permits the use of constellations having any integer number of points per ring, and a varying number of rings from symbol time to symbol time.

Description

SIGNALLING METHOD USING MULTIPLE MODULUS CONVERSION AND SHELL MAPPING

COPYRIGHT NOTICE

A portion of the disclosure of this patent document contains material that is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.

BACKGROUND OF THE INVENTION A. Field of the Invention

The present invention relates to a method and device for selecting data symbols to be transmitted in a digital communication system. Although the invention provides a method of signal point mapping and may be applied to any data communication system, it is particularly well suited for use in systems where the selection of constellation points is subject to relatively strict criteria, e.g., where signal points within a constellation are not equally spaced, or where a constellation or constituent sub-constellations may contain a number of points that is not a power of two. The method described herein is also well suited for the case where the number of elements within the constellations varies from symbol to symbol. Such restrictions are encountered in communication systems that use the public digital telephone network where one side of the communication link (typically a server) has direct digital access to the network. In such a system the constellation selection is limited in part by the presence of Robbed-Bit-Signaling (RBS) and/or a Network Digital Attenuators (NDA), and power constraints imposed by government (e.g., FCC) regulations. These techniques also can be used to help minimize the impact of RBS/NDA on the data rate over such channels. B. Description of the Related Art

As stated above, the signaling method and device described herein may be used in many types of digital communication systems. Generally, modulation techniques for the transmission of digital information involve modulating the amplitude and phase of a carrier frequency. A baseband signal (an unmodulated information sequence such as a train of pulses of various amplitudes) may be used to modify the amplitude of a carrier frequency sine wave. Because a carrier may also be separated into orthogonal cosine and sine components (also referred to as inphase (I) and quadrature (Q) channels), a modulated carrier may be thought of as the sum of a modulated sine wave and a modulated cosine wave.

As is well known in the art, a two-dimensional plane, or I-Q plane, is used as a shorthand notation to represent the amplitude and phase of the carrier. The signals that make up a signal constellation are represented as points in the I-Q plane, which are usually set out in a grid-like fashion. A particular signal point may be specified as a coordinate pair in the I-Q plane. The points in the I-Q plane are also generally referred to as a baseband representation of the signal because the points represent the amplitudes by which the sine and cosine components of a carrier will be modified. Each "signal point" is also referred to herein as a "symbol."

While the invention described herein is applicable to systems that use modulated carriers as described above, the preferred embodiments are essentially baseband systems that do not involve the modulation of a carrier. Consequently, the signal points are selected from a single-dimensional signal space, as opposed to a two- dimensional inphase/quadrature signal space. The system for which the invention is particularly well suited uses the public digital telephone network. 1. Digital Telephone Network

For many years the public digital telephone network (DTN) has been used for data transmission between modems. Typically, a modulated carrier is sent over a local loop to a service provider (e.g., a Regional Bell Operating Company), whereupon the service provider quantizes the signal for transmission through the DTN. A service provider that is located near the receiving location converts the digital signal back to an analog signal for transmission over a local loop to the receiving modem. This system is limited in the maximum achievable data rate at least in part by the sampling rate of the quantizers, which is typically 8 kHz (which rate is also the corresponding channel transmission rate, or clock rate, of the DTN).

Furthermore, the analog-to-digital (A/D) and digital-to-analog (D/A) conversions are typically performed in accordance with a non-linear quantizing rule.

In North America, this conversion rule is known as μ-law. A similar non-linear

sampling technique known as A-law is used in certain areas of the world such as Europe. The non- linear A/D and D/A conversion is generally performed by a codec (coder/decoder) device located at the interfaces between the DTN and local loops.

It has been recognized that a data distribution system using the public telephone network can overcome certain aspects of the aforesaid limitations by providing a digital data source connected directly to the DTN, without an intervening codec. In such a system, the telephone network routes digital signals from the data source to a client's local subscriber loop without any intermediary analog facilities, such that the only analog portion of the link from the data source to the client is the client's local loop (plus the associated analog electronics at both ends of the loop). The only codec in the transmission path is the one at the DTN end of the client's subscriber loop.

Figure 1 shows a block diagram of a data distribution system. The system includes a data source 10, or server, having a direct digital connection 12 to a digital telephone network (DTN) 14. A client 16 is connected to the DTN 12 by a subscriber loop 18 that is typically a two-wire, or twisted-pair, cable. The DTN routes digital signals from the data source 10 to the client's local subscriber loop 18 without any intermediary analog facilities such that the only analog portion of the link from the server 10 to the client 16 is the subscriber loop 18. The analog portion thus includes the channel characteristics of the subscriber loop 18 plus the associated analog electronics at both ends of the subscriber loop 18. The analog electronics are well known to those skilled in the art and typically include a subscriber line interface card at the central office that includes a codec, as well as circuitry used to generate and interpret call progress signals (ring voltage, on-hook and off-hook detection, etc.). In the system of Figure 1 , the only D/A converter in the transmission path from the server 10 to the client 16 is located at the DTN 14 end of the subscriber loop 18. It is understood that the client-side, or subscriber-side, equipment may incorporate an A/D and D/A for its internal signal processing, as is typical of present day modem devices. For the reverse channel, the only A/D converter in the path from the client 16 to the server 10 is also at the DTN 14 end of the subscriber loop 18.

In the system of Fig. 1, the server 10, having direct digital access to the DTN

14 may be a single computer, or may include a communications hub that provides digital access to a number of computers or processing units. Such a hub/server is disclosed in U.S. Patent No. 5,528,595, issued June 18, 1996, the contents of which are incorporated herein by reference. Another hub/server configuration is disclosed in U.S. Patent No. 5,577,105, issued November 19, 1996, the contents of which are also incorporated herein by reference.

In the system shown in Fig. 1, digital data can be input to the DTN 14 as 8-bit bytes (octets) at the 8 kHz clock rate of the DTN. This is commonly referred to as a DS-0 signal format. At the interface between the DTN 14 and the subscriber loop 18, the DTN 14 codec converts each byte to one of 255 analog voltage levels (two different octets each represent 0 volts) that are sent over the subscriber loop 18 and received by a decoder at the client's location. The last leg of this system, i.e., the local loop 18 from the network codec to the client 16, may be viewed as a type of baseband data transmission system because no carrier is being modulated in the transmission of the data. The baseband signal set contains the positive and negative voltage pulses output by the codec in response to the binary octets sent over the DTN. The client 16, as shown in Figure 1, may be referred to herein as a PCM modem.

Figure 3 shows a μ-law to linear conversion graph for one-half of the μ-law

codeword set used by the DTN 14 codec. As shown in Fig. 3, the analog voltages

(shown as decimal equivalents of linear codewords having 16 bits) conesponding to the quantization levels are non-uniformly spaced and follow a generally logarithmic curve. In other words, the increment in the analog voltage level produced from one codeword to the next is not linear, but depends on the mapping as shown in Fig. 3. Note that the vertical scale of Fig. 3 is calibrated in integers from 0 to 32,124. These numbers conespond to a linear 16-bit A/D converter. As is known to those of ordinary skill in the art, the sixteenth bit is a sign bit which provides integers from 0 to -32124 which correspond to octets from 0 to 127, not shown in Fig. 3. Thus Fig. 3 can be viewed as a conversion between the logarithmic binary data and the corresponding linear 16-bit binary data. It can also be seen in Fig. 3 that the logarithmic function of the standard conversion format is approximated by a series of 8 linear segments.

The conversion from octet to analog voltage is well known, and as stated

above, is based on a system called μ-law coding in North America and A-law coding

in Europe. Theoretically, there are 256 points represented by the 256 possible octets,

or μ-law codewords. The format of the μ-law codewords is shown in Fig. 2, where

the most significant bit b₇ indicates the sign, the three bits b₆-b₄ represent the linear segment, and the four bits, b₀-b₃ indicate the step along the particular linear segment. These points are symmetric about zero; i.e., there are 128 positive and 128 negative levels, including two encodings of zero. Since there are 254 non-zero points, the maximum number of bits that can be sent per signaling interval (symbol) is just under

8 bits. A μ-law or A-law codeword may be referred to herein as a PCM codeword. It

is actually the PCM codeword that results in the DTN 20 codec to output a particular analog voltage. The codeword and the corresponding voltage may be referred to herein as "points."

Other factors, such as robbed-bit signaling, digital attenuation (pads), channel distortion and noise introduced by the subscriber loop, and the crowding of points at the smaller voltage amplitudes and the associated difficulty in distinguishing between them at the decoder/receiver, may reduce the maximum attainable bit rate. Robbed Bit Signaling (RBS) involves the periodic use of the least significant bit (LSB) of the PCM codeword by the DTN 14 to convey control information. Usually the robbed bit is replaced with a logical '1' before transmission to the client 16. In addition, due to the fact that a channel might traverse several digital networks before arriving at the terminus of the DTN 14, more than one PCM codeword per 6 time slots could have a bit robbed by each network, with each network link robbing a different lsb.

To control power levels, some networks impose digital attenuators that act on the PCM codewords to convert them to smaller values. Unlike most analog attenuators, a network digital attenuator (NDA) is not linear. Because there are a finite number of digital levels to choose from, the NDA will be unable to divide each codeword in half. This causes distortion of the analog level ultimately transmitted by the codec over the subscriber loop 18. RBS and an NDA can coexist in many combinations. For example, a PCM interval could have a robbed bit of type '1', followed by an NDA followed by another robbed bit of type '1'. This could happen to a byte if a channel goes through a bit-robbed link, then through an NDA, then another bit-robbed link before reaching the DTN 14 codec.

It is evident that the above-described data transmission system imposes many constraints on the points that may be used to form a signal constellation. Inter alia, the octets will always be converted to non-linearly spaced voltage pulses in

accordance with μ-law or A-law conversion; lsbs may be robbed during some time

slots making some points unavailable in that time slot; digital attenuators may make some points ambiguous; and noise on the local loop may prevent the use of closely spaced points for a desired error rate.

2. Shell Mapping

Shell mapping is a prior art technique of assigning transmit data bits to constellation points, or symbols. Shell mapping requires that an encoder group symbols together in mapping frames, typically consisting of eight symbols. The constellation is divided into a set of rings that are generally concentric, with an equal

1 number of points per ring. A block of transmit data bits is then used to select a sequence of ring indices and to select the points to send from each of the selected rings.

For purposes of determining the most desirable sequence of rings to select, each ring is assigned a cost value associated with its distance from the origin. The costs are generally representative of the power needed to transmit a point from within that ring. For each possible sequence of rings, a total cost is calculated. The ring index selector/encoder is designed to prefer the least-cost ring sequences, i.e., low- powered point sequences are preferred over high-power point sequences. Note that the above scheme of eliminating certain ring sequences is only possible when there is signal set redundancy. This requires the constellations to be larger than would otherwise be required in systems that do not utilize shell mapping. Furthermore, because the excluded ring sequences are those that include large numbers of outer rings and the favored ring sequences are those that include large numbers of inner rings, the transmitted constellation points will have a non-equi-probable distribution. As a result, the shell mapping technique tends to reduce average power for a given spacing between points (called o _,,,,), thereby increasing the peak-to-average power ratio (PAR).

The reduction in average power (compared to an equi-probable point distribution) can be exploited in an average power constrained system by increasing the distance between points until the average power is equal to an equivalent system with equi-probable point distribution. The receiver sees points with improved d_^,, that are easier to detect in the presence of noise. Generally, the possible improvement is less than 2.0 dB, and the practically realizable benefit is about .5 to 1.0 dB. The advantage of shell mapping applies to a system where the points are equally distributed within the signal space. This means that the spacing between points in the center of the constellation is the same as the spacing between points at the outer edge. This is not the case in PCM modems where the constellation points are inherently unequally spaced. There is a strong incentive to minimize constellation size in such a scenario. As shown in Figure 3, the codes in a PCM codec are arranged in segments of 16 members each, with each higher amplitude segment having twice the spacing of the next lower segment. Constellation expansion involves adding more low-power points that are spaced closer together than high-power points. In some cases, the addition of a single point can reduce minimum spacing dramatically. Thus the idea of expanding the constellation to achieve better cL^, may have the opposite effect, and reduce it.

It is clear that traditional shell mapping has disadvantages when utilized in a system having constraints on signal point selection that are inherent in PCM modems. Described herein is an improved signal mapping method that includes features of Multiple Modulus Conversion to obtain an effective signal mapping technique.

°\ SUMMARY OF THE INVENTION

The method of the present invention uses Multiple Modulus Conversion (MMC) to map data bits to signal points (or symbols) of a signal set. This is also referred to as Mixed-Base Mapping (MBM). One preferred embodiment maps bits directly to a frame of symbols using MMC. The MMC apparatus converts blocks of data bits to a corresponding block of ,-ary, ₂-ary,..., _n symbols. A subset of codec codewords is used to represent the -ary signals. For each time slot (symbol time) one of M separate octets are selected for transmission by the encoder, and the encoder's output is sent through the DTN to a subscriber loop codec. The value of M can vary among the different time slots. The analog output of the codec corresponds to -level, or M-ary pulse amplitude modulation, because each of the transmitted octets is converted to one of M analog voltages at the DTN's codec before being communicated over the subscriber loop. The set of M levels is also referred to as a signal constellation, with the M different voltage levels referred to as signal points within the constellation. Another preferred method, referred to as Multiple Modulus Shell Mapping (MMSM), provides a combined frame mapping technique that uses MMC and Shell Mapping (SM) to map data bits to a sequence of data symbols, or points. The MMSM apparatus includes a traditional shell mapper to generate ring indices from K data bits, and a modulus converter to select the signal points from within the ring based on B data bits. MMSM permits the use of constellations having any integer number of points per ring. MMSM also accommodates variations in the constellations from time-slot to time-slot within a frame. That is, the signal set for each symbol time within a frame may have a different number of points per ring, while the number of rings in each constellation remains constant. MMSM produces d_m,,, equal to the best of MMC and SM, and in some cases c _^,, may be better than that for either MMC or SM. A further embodiment generates ring indices by using MMC rather than traditional ring generation methods. This is referred to herein as Mixed Base Ring Selection (MBRS). The Mixed-Base Ring Selector permits the selection of constellation sets where the number of rings may vary from symbol time to symbol time. This further increases the flexibility of the signal set selection. The signal points are selected from within a chosen ring based on straightforward bit mapping, or mapping based on well-known techniques of trellis coding. Alternatively, MBRS may be used to generate the ring indices in a MMSM implementation, such that both the number of rings and points per ring vary from symbol time to symbol time (note that the number of points per ring is constant in a given symbol time).

// BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other objects, features and advantages of the present invention will be more readily appreciated upon reference to the following disclosure when considered in conjunction with the accompanying drawings, in which: Figure 1 depicts a communications network with a data source having direct digital access to the DTN;

Figure 2 shows the elements of a μ-law codeword;

Figure 3 shows a μ-law to linear conversion graph;

Figure 4 shows a flow chart of one embodiment of Multiple Modulus Conversion;

Figure 5 shows a Multiple Modulus Converter;

Figure 6 shows a block diagram of a preferred embodiment of the Multiple

Modulus Shell Mapper; and,

Figure 7 shows a block diagram of an alternate preferred embodiment of the Multiple Modulus Shell Mapper.

I d'- DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

Multiple Modulus Conversion

In order to provide flexible data rates, a preferred embodiment provides a method of sending a fractional number of bits per symbol, so as to achieve maximum data rate and minimum required SNR. In addition, it is desirable to mitigate the effects of robbed-bit signaling (RBS), such that the highest possible data rate is achieved. Both these requirements can be met with multiple modulus conversion (MMC).

Constellations consisting of M_t points are selected to meet the following three criteria:

1. 2* < ,. ,

where K is the number of user bits to be transmitted in an n symbol frame.

2. Points in constellations used in RBS intervals, should be chosen such that the

least significant bits (lsbs) in their corresponding μ-law codes are least effected by

RBS. (Note these constellations will always be smaller than the ones used in non-rbs intervals, since less data is sent during this time).

3. The probability of symbol error is minimized and balanced. Minimizing the probability of symbol error is accomplished by choosing points with maximal spacing, while constraining the points to have some specified average power. In addition, the number of points with minimum spacing should be minimized (smallest number of nearest neighbors). Balancing the symbol error probability, means that mutually, the M_i should have about the same symbol error rate, since then the overall error rate (frame or block error rate, for example) will not be dominated by the symbol error rate of any one constellation.

\ 'b Figure 5 depicts a Multiple Modulus Converter. The encoder first collects K bits to send during the n symbol frame. The K bits are mapped to n indices, representing R7n bits/symbol (not necessarily an integer). One technique of implementing the conversion process is to represent the K bits as an integer, I, where:

I = b₀ +b 2 + b_i2² +...+b_κ._λ2^κ' (_eq. 1)

In eq.1 , b₀ is the lsb and b_κ_ is the msb (most significant bit) of the R^"-bit data block.

The number / may also be represented in a mixed-base, or multiple modulus, format such as:

I = m + m₂M^ + m₃M_lM₂ + ... + m_nM_lM₂...M_n__] .

The result of the modulus conversion process is the production of the values

m. ,i = l,.-,n , where the values of the m, are constrained to the interval: 0 < , < M₍ .

The n values of m, correspond to codec codewords that will be transmitted through the

DTN in digital format, and ultimately converted to analog levels by the DTN D/A codec for transmission to the subscriber. A simple algorithm can be used for generating the m„ given the M, as follows:

For i=l,...,n

I = I/M, Q = Int(I) F = I- Q BI, = M*F ι = Q

Next i

The m, are indices to symbols that are conveyed by the encoder/server to the receiver by transmitting to the DTN the codewords corresponding to the appropriate points. The indices may be generated in other fashions, using similar algorithms, without departing from the spirit and scope of the invention.

11 For example, given K bits in decimal equivalent format, represented as ₀, an equivalent expression to generate the indices is given by:

where 0 < j ≤ F - 2 , x_J+] = x _! ~ ^m, , and m_ΕΛ~x_ΕΛ, the C, are the moduli from

symbol time to symbol time, and F-1 is the number of indices generated (typically, F-

1 = 6).

Figure 4 depicts the steps performed in converting the data bits to symbols. At step 20 the K bits are collected for transmission through modulus conversion. At step

22 the bits are converted to an integer value. At step 24 the integer is converted to multiple modulus indices by using the moduli M, values. At step 26, the multiple modulus indices are transmitted to the telephone network in the form of the corresponding codec codewords.

At the decoder, the process of decoding or recovering the data involves identifying the correct symbol indices m, from the analog voltages sent over the receiver's subscriber loop and then unmapping these indices through Reverse Multiple

Modulus Conversion. One preferred method of decoding the symbols is to first restore the integer / from the recovered indices m The following algorithm may be used:

1 = 0

For i = l,...,n I = M„__l+1*I + m_n__l+1

Next I

From I, the K bits in the frame can be recovered.

i According to the preferred embodiment of the present invention, the values of the _j (or and their corresponding constellation sets are ultimately determined in large part by the SNR of the channel. One of ordinary skill in the art can appreciate that the SNR can be determined in a number of ways, for example during an initialization period, a sequence of known codewords may be transmitted to produce a known sequence of symbols by the DTN's codec, and the variance from the expected symbols (points) is measured. Once the SNR is determined, a search method can be employed to determine the M, 's and their corresponding constellations. The constellations are chosen from the DAC codes that satisfy the three criteria cited above. That is, search for n sets of points, corresponding to n symbols per frame, and having n moduli, which simultaneously minimizes the desired probability of symbol error, while satisfying the data rate criterion number 1 , above.

One can use the following expression for a good approximation to the probability of symbol eπor when performing the constellation search:

where

^&x)=τ≡Y'"^ά'-

N_A is the number of points with minimum spacing Δ , and σ is the noise standard

deviation. Also, one needs to account for the number of points with spacing Δ , which can be handled as a multiplicative factor times the probability of error.

One advantage of MMC is that it allows a non-integer number of bits to be mapped to each symbol, which increases efficiency, because the constellation sizes are not restricted to powers of two (i.e, a fractional number of bits/symbol are allowed). As a specific example, in a system having one RBS link, the block of data

lb is converted to a multi-digit modulo-M, number (where each digit represents an M,- ary symbol) followed by a single digit modulo-M₂ number (representing an additional M₂-ary symbol). The concatenated symbols, or digits of the modulo-M, and modulo- M₂ numbers, are then transmitted in binary format as eight-bit bytes, or octets, through the telephone system where each octet represents a symbol. At the local loop on the receiver end of the link, the telephone system converts the octets to analog voltages corresponding to the M,-ary and M₂-ary symbols. The block of symbols may then be decoded by a reverse modulus conversion to recover the binary data.

Figure 5 depicts a modulus converter 30 for encoding information bits for transmission. The converter includes an input buffer 32 for accepting a block of data bits received on line 34, and at least one modulus buffer 36 for storing the modulus values 38. Alternatively, multiple buffers may be provided in the case of multiple modulus conversion. A processing unit 40 is connected to the input buffer 32 and the modulus buffer(s) 36. The processing unit 40 converts the data bits to symbols using the modulus value(s) 38. The converter 30 then outputs the symbols to the telephone system on output 42. The output 42 is preferably a port of a microprocessor. The modulus converter preferably has six modulus buffers, one for each modulus value (corresponding to each time slot of a six-slot frame). The modulus converter's processing unit 40 is preferably a microprocessor, or digital signal processor. The buffers 32 and 36 are preferably random-access memory (either on-chip, or in associated RAM chips). Alternatively, the conversion function may be implemented in an application specific integrated circuit, having optimized conversion logic and hardware buffers.

As a simple example, in a system having one RBS link in the sixth time slot of a 6 slot frame, the block of data is converted to a multi-digit modulo-M, number

11 (where each digit represents an M,-ary symbol) followed by a single digit modulo-M₂ number (representing an additional M₂-ary symbol). The concatenated symbols, or digits of the modulo-M, and modulo-M₂ numbers, are then transmitted in binary format as eight-bit bytes, or octets, through the DTN 14 where each octet represents a symbol. At the local loop 18 on the receiver end of the link, the DTN 14 codec converts the octets to analog voltages corresponding to the M,-ary and M₂-ary symbols. The block of symbols may then be decoded at the client 16 by a reverse modulus conversion to recover the binary data.

The MMC method therefore provides a general method for mapping data to a multi-symbol number having various moduli that allows for tighter constellation packing and constellation balancing. As a result, a minimum number of constellation points are required for a given bit capacity, and the error rate is not dominated by any one symbol interval. Furthermore, MMC lowers the required SNR to achieve a desired error rate. Multiple Modulus Shell Mapping

A preferred Multiple Modulus Shell Mapping device, in accordance with a further embodiment of the present invention, is shown in Figure 6. The MMSM device includes a bit parser 100, modulus converter 110, shell mapper 120, point selectors 130-136, and parallel to serial converter 140. The input data bits are scrambled, as is known in the art, to ensure bit transitions even if the actual data is an all zero data stream. Bit parser 100 divides the bits into modulus mapping bits b₀ through b_B_, and shell mapping bits SL_Q through a_κ_,. As described below, the modulus converter 110 converts the B bits b₀ - b_B_, into N (six) multiple modulus coefficients m₀ through m₅, based on the predetermined moduli M, through M_N. The shell mapper 120 maps the K bits ά_B through ά_B+κ_, to ring indices r₀ through r₅ based on a table look-up or, preferably, on well known mapping algorithms. The ring indices r₀ through r₅ take on integer values from 0 to R-l, where R is the number of rings. The point selectors 130-135 determine the actual point to be transmitted based on inputs from the modulus converter 110 and shell mapper 120, and provide the corresponding PCM codewords at their outputs. Parallel to serial converter 140 outputs the PCM codewords in a serial fashion. The preferred MMSM operates on a data frame of N=6 symbols. Other frame sizes may alternatively be used. The transmitter and receiver during an initiation process negotiate the remaining parameters K, B, M„ and R.

Figure 7 depicts an alternative preferred embodiment of the Multiple Modulus Shell Mapper that includes sign mapper 260, differential sign bit encoder 250, PCM codeword-to-linear codeword converter 270, and integrator 280. Instead of using all B bits in the modulus converter 210, only D bits are used, where D=B-N+1. The outputs of the point selectors provide unsigned points U₀ through U₅ (i.e., codewords containing the seven least significant bits, without the sign bit). In this description, N is again assumed to be six. Other frame sizes may alternatively be used.

The remaining N-l bits are designated s₀ through s^, and are differentially encoded into bits ds₀ through ds^, by differential sign bit encoder 250. The output of differential sign bit encoder 250 is sent to sign mapper 260. Sign mapper 260 uses the N-l bits (which is preferably five bits in a frame of six symbols) to assign sign bits to five of the six unsigned points U₀ through U₅. The five lowest magnitude unsigned points are given sign bits in this manner. Sign mapper 260 also provides DC compensation by providing a sixth sign bit based on the DC offset signal provided by the running sum integrator 280. The sign mapper 260 inserts a sign bit in the largest magnitude unsigned point (or first largest in the event that two are equal magnitude)

ή that is opposite the sign of the DC offset. DC offset will tend to be decreased in this manner.

Alternative DC compensation schemes may be used that insert a compensation sign bit every other frame, or 1 bit in every three frames, etc. Furthermore, schemes may be used that insert a compensation bit in every time slot of every frame. Further details of the DC compensation aspect of the sign mapper are described in co-pending application, serial number 08/871,220, filed June 9, 1997, entitled Frame-Based Spectral Shaping Method And Apparatus the contents of which are hereby incorporated herein by reference. MMSM Ring Sequence Selection

To select the N ring indices r„ the improved mapping method and apparatus uses certain aspects of the prior art shell mapping devices. The sequence of rings may be stored in a lookup table or may be derived from well-known shell mapping algorithms. One such algorithm is set forth in Recommendation V.34, an international standard set forth by the International Telecommunication Union.

Traditional shell mapping requires a mapping frame consisting of N symbols into which K+B bits will be mapped. The preferred data frame size is N=6, although other frame sizes may be used. The frame size of six is desirable because robbed bit signaling typically occurs every sixth octet of a DS-0 channel. The constellation is divided into R rings with M, points per ring. In every frame a data parser allocates K bits per frame to ring selection and B bits per frame to point selection in the base rings. The traditional shell mapper uses the K bits to

generate N ring indices r_h 0 < r, < R using an algorithm for selecting r_t such that there

is a unique mapping between the K input data bits and the sequence of r,. The K bits are recoverable from the ring indices r, by a similar algorithm at the decoder. The possible combinations of K bits is 2* and the number of possible combinations of rings

is R", so this imposes the restriction that 2^K < R^N, however, for most cases R" will be

strictly larger than 2^K to provide signal set redundancy for shaping gain.

Set forth below is a program that may be used to generate the shell mapping indices in accordance with the preferred frame duration of six time slots and R = 9 rings. The program also includes the decoder algorithm. The program is written in the Matlab language, which is widely used in the engineering community. The reader should be aware that the choice of variable names are not identical to those defined above; however, the variable assignments will be clear from the context of the program.

% smδ . m

% This program performs shell mapping and inverse mapping.

% The frame size is 6. % The ring indices will be contained in R6.

% clear all;

M = 9; % Number of rings

Xstart = 300000 ;%64317;%240000 % Start of data to be % shell mapped err=0;

N=l;%36000; % Number of Ring indices to compute

R6 = zeros (N, 6); % Ring indices

% Get z6 gl = [ones(M,l) ' zeros (M-l, 1) ']' ; g2 = conv(gl,gl) g3 = conv(gl,g2) g6 = conv(g3,g3) z6 = zeros (length (g6) , 1) ; for i = 1 : length (g6) , z6(i+l) = sum(g6 (0+1: (i-l)+l) ) ; end; for n=l:N, % Start ring indices loop

X = Xstart+n-1;

% Find the index, i, in z6, s.t. z6 (i) <= X i=0; while (X >= z6(i+l) ) i = i+1; end; i=i-l; % Adust i due to while loop

% Compute the residue, r6 r6 = X - zδ(i+l) ;

% Now compute the ring indices

*l % r6 z = r6;

S=-l; while (z >= 0) s = s+1; z = z - g3 (s+1) *g3 ( (i-s)+l) ; end;

% At this point s and (i-s) remain fixed

% r31 z = z + g3 (s+1) *g3 ( (i-s)+l) ; r3 = z ; r31 = rem(r3,g3 (s+1) ) ; z = r31; j =-l; while (z >= 0) % s is fixed and j=0,l,...,s j = j+l; z = z - g2 ( (s-j)+l) ; end; z = z + g2 ( (s-j) +1) ; r21 = z;

% Get the ring indices (R1,R2,R3)

R6(n,l) = j; if ( (s-j) <= M-l)

R6 (n,2)=r21; R6 (n,3)=(s-j) -r21; ; else

R6(n,2)=(s-j) -(M-l) + r21; R6 (n, 3 ) = (M-l) -r21; end; % r32 r32 = floor (r3/g3 (s+1) ) ; z = r32; j =-l; while (z >= 0) % (i-s) is fixed and j=0, 1, ... , (i-s) j = j+ ; z = z - g2 ( ( (i-s) -j)+l) ; end; z = z + g2 ( ( (i-s) -j)+l) ; r22 = z;

% Get the ring indices (R4,R5,R6) R6(n,4) = j; if ( (i-s) -j <= M-l)

R6 (n,5)=r22; R6(n,6)=( (i-s) -j) -r22; ; else

R6(n,5)=((i-s)-j)- (M-l)+r22; R6 (n, 6) = (M-l) -r22 ; end; % *** Decode = inverse shell mapping

% Get rr31 ss = sum(R6 (n, 1 :3) ) ; jj = R6(n,l) ; if ((ss-jj) <= (M-l)) rr21 = R6 (n,2) ; else rr21 = (M-l) -R6 (n,3) ; end; rr31=rr21; for k=0:jj-l, rr31 = rr31 + g2 (ss-k+1) ; end;

% Get rr32

) itø- rr22 = R6 (n,5) ; else rr22 = (M-l) -R6(n,6) ; end; rr32=rr 2; for k=0:jj-l, rr32 = rr32 + g (ss-k+1) ; end;

% Finish decode = inverse SM ii = sum(R6 (n, : ) ) ;

% Get rr6 rr6 = rr32*g3 (ss+1) + rr31; for k=0:ss-l, rr6 = rr6 + g3 (k+1) *g3 (ii-k+1) ; end;

XX = z6 (i+1) +rr6; % Decoded data if (XX ~= X) err = err+1; end; if (mod (X, 1000)== 0)

X err end; end; % End data loop

Multiple Modulus Conversion (MMC), without shell mapping, has been described above. Generally, Multiple Modulus Conversion provides a method of converting blocks of binary data to a corresponding block of M,-ary, M₂-ary,...,M_n symbols to maximize the data rate while minimizing the required Signal-to-Noise Ratio (SNR) to achieve a desired error rate. The techniques of MMC are combined with traditional Shell Mapping to achieve superior performance in the form of MMSM.

In prior art shell mappers, Qi bits are used to select the point from the designated ring , with each ring containing 2^Ql points. The following relationship must be met in prior art shell mappers:

i

In contrast to the prior art methods, the present invention uses a multiple modulus conversion technique to select the points in the base ring. Multiple Modulus

Conversion removes the 2^Qt points per ring restriction associated with prior art shell mapping techniques and allows any integer number of points per ring. There can be up to N unique moduli M, per frame of length N. The moduli are held to the less strict requirement that the product of moduli is greater than or equal to 2^B, i.e.,

^{2* ≤} π, > ^(eq- 2⁾ where B bits per frame are allocated to multiple modulus conversion for the purpose of assigning points to the base ring, and i is an index for the slot within the frame.

In accordance with the preferred embodiments, the constellations to be used are agreed upon in a negotiating process between the transmitter and receiver based upon the level of noise and upon network impairments such as RBS and ΝDAs. The number of rings R (and hence the moduli M,) preferably are also agreed upon. Each constellation is equally divided into R rings where M, is the number of points per ring for the time slot i within the frame. In the system of Figure 6, each of the rings includes positive PCM codewords (or points) and the corresponding negative PCM codewords (or points). Note that in the system of Figure 7, the modulus M. is equal to the number of positive points (or negative points) because the sign is assigned after the modulus conversion. The use of zero-valued codewords can also be accomodated. The modulus converter collects K bits to send during the N symbol frame. The following process allows theR^" bits to be mapped to N indices, m

As in the basic MMC process, the next step of the conversion process is to represent the K bits as an integer, J where:

/ = b₀ + b_λ 2 + b₂2² + ... + b_B_.2^B~ (eq. 3)

In eq. 3, b₀ is the lsb (least significant bit) and b_B_, is the msb (most significant bit) of

the R-bit data block. The number /may be also be represented as: I = m_λ + m₂M_l + m₃M ₂M ₃ + ^■ • • + m_NM_xM₂ ■ ■ ■ M _N__λ (eq. 4)

The result of the modulus conversion process is the production of the values

m_l ,i = \,...N, where the values of the m, are constrained to the interval 0 < m, < M, .

The N values of TM, generally correspond to N analog levels that will be produced by the network's codec. The moduli M. may be referred to herein as "base moduli" to distinguish them from the "moduli indices," m„ which are the coefficients of the base moduli in eq. 4.

A simple algorithm can be used for generating the m„ given the M„ as follows:

Forj=l,...,N

I = I/M_t Q = Int(I) F = I- Q m_l = M*F I = Q

Nextj

The m_l are indices to symbols that are conveyed by the encoder/server to the receiver by transmitting to the DTΝ the codewords corresponding to the appropriate points. At the decoder, the process of decoding or recovering the B data bits involves identifying the correct symbol indices m_l from the analog voltages sent over the receiver's subscriber loop and then unmapping these indices through Reverse Multiple Modulus Conversion. In MMSM, the values of the , are determined based only upon the point within the ring, and not the ring itself. The ring indices are decoded separately, as discussed above.

One method of decoding the symbols is to first restore the integer / from the recovered indices m The following algorithm may be used:

^<±£ 1 = 0

Forj = E....N

/ = M_N__J+l*I + m_N__j+1 Nextj From I, the B bits in the frame can be recovered in a straightforward manner.

Note that in the alternative preferred embodiment of Figure 7, a subset of the B bits are used by the modulus converter 210. These bits are converted using the above- described algorithm, with the understanding that the moduli M, through M_N are selected appropriately. Because the sign of the PCM codeword is assigned after the modulus conversion of converter 210, the M,-M_N of Figure 7 are one half the values of an exactly equivalent system of Figure 6. Obviously an exact equivalent will not exist when any of the moduli of Figure 6 are odd - this observation is made only to clarify the relationship between the systems of Figures 6 and 7. In practice, if DC compensation is implemented as in Figure 7, optimal values oϊM,-M_N are chosen for that scenario.

The parameters needed to describe the mapping algorithm for the transmitter are K ring selection bits, B point selection bits are used for selecting the points from within the rings based on the multiple modulus conversion, R is the number of rings within the constellation, and M, are the moduli for each time slot denoted by subscript i. By making .KM) (or R=l) the mapping scheme collapses to MMC. By making M=2^β' the mapping scheme collapses to SM. In some instances (e.g., for a given constellation) collapsing to one scheme or the other provides the best results. In most cases however, the MMSM technique gives the best result.

A mapping scheme has been described that combines shell mapping and multiple modulus conversion. The combination collapses to either scheme at the discretion of the receiver. This allows the receiver total flexibility to be able to achieve the largest d_mιη at all data rates. The complexity increase in the transmitter is small compared to that of straight shell mapping. The complexity increase over multiple modulus conversion is significantly greater. In the receiver there is little complexity increase due to the decoder. However, as the choice of mapping lies with the receiver, the receiver does not have to accept any additional complexity due to this scheme.

Mixed-Base Ring Selection

The Mixed Base Ring Selector uses MMC to select ring indices. The rings are chosen based upon the indices generated by the Multiple Modulus Conversion of K bits. That is, the MBSR operates in the same manner as the Multiple Modulus Converter of Figure 5, but is used to generate ring indices instead. The MBSR may then be used in the embodiments of Figures 6 and 7 in place of ring generator 120 or 220. In this way, the number of rings may vary from time slot to time slot, and the number of points per ring may vary from time slot to time slot. However, the number of points per ring within a specific time slot must be constant. Alternatively, another preferred embodiment utilizes the MBRS (to generate the ring indices) in combination with traditional uncoded bit mapping techniques (i.e., each point in a given ring represents a particular bit sequence). Yet another preferred embodiment utilizes the MBRS in combination with well-known trellis coding techniques. Preferred embodiments of the present invention have been described herein. It is to be understood, of course, that changes and modifications may be made in the embodiment without departing from the true scope of the present invention, as defined by the appended claims.

^

Claims

We claim:

1. A method for transmitting digital information from a data source having direct access to a digital telephone system for supplying binary codewords directly to the telephone system, to a receiver connected to said telephone system by an analog subscriber line, wherein the telephone system converts the binary codewords from the data source to analog voltage levels for transmission to the receiver, the method comprising the steps of: selecting a number of symbol periods, n, in a frame; selecting at least one block of information bits to be transmitted in said frame, wherein said at least one block comprises K bits; selecting at least one set of symbols for each of said n symbol periods wherein the number of symbols in each said set corresponds to a modulus M of said symbol period, and wherein each said symbol corresponds to a binary codeword; mapping said at least one block of information bits to said symbols by multiple modulus conversion; and providing said codewords to the telephone system.

2. The method of claim 1 wherein said modulus M varies between at least two of said symbol periods.

3. The method of claim 1 wherein n is a multiple of 6.

4. A method of encoding information bits for transmission over a digital telephone system where an encoder has direct digital access to the telephone system, and wherein said telephone system appropriates bits during known bit-robbed rate time slots, comprising the steps of: providing desired moduli values, M,, M₂,...,M_n for n time periods; providing n sets of binary codewords that define n sets of symbols;

9^ mapping information bits to said binary codewords selected from said n sets of symbols using said moduli values; and providing said codewords to said telephone system.

5. The method of claim 4 where the moduli values of M,, M₂,... _n are such that n [ M_; ΓëÑ2^K, where K is the number of bits converted to symbols. ╬╣=l

6. A method of decoding data symbols transmitted over a digital telephone system where an encoder has direct digital access to the telephone system for providing binary octets to the telephone system, and wherein said telephone system appropriates bits from said octets during known bit-robbed time intervals, and wherein said telephone system converts the octets to analog voltages, comprising the steps of: selecting n sets of symbol codewords; receiving analog voltages representing symbols to be decoded; and converting said received symbols to a block of binary data using multiple modulus conversion.

7. A modulus converter for encoding information bits for transmission over a digital telephone system where an encoder has direct digital access to the telephone system, and wherein said telephone system appropriates bits during known bit-robbed rate time slots, comprising: an input buffer for accepting a block of data bits; at least one modulus buffer for storing at least one modulus value; a processing unit connected to said input buffer and said at least one modulus buffer, for converting said block of data bits to symbols using said at least one modulus value; and an output connected to said processing unit for communicating said symbols to

9A the telephone system.

8. The modulus converter of claim 7 wherein said at least one modulus buffer comprises six modulus buffers, wherein each of said six modulus buffers is adapted for storing a corresponding modulus value.

9. The modulus converter of claim 7 wherein said processing unit is a microprocessor.

10. The modulus converter of claim 7 wherein said input buffer is a random- access memory register.

11. A method of mapping data bits to a sequence of signal points selected from a signal constellation wherein the points in the signal constellation are equally divided into rings, the method comprising the steps:

segmenting a plurality of data bits into a first block and a second block;

selecting a sequence of ring indices in response to said first block;

determining a sequence of moduli indices m_; in response to said second block wherein at least one of said moduli indices has a base modulus that is a value other than a power of two; and

selecting a signal point sequence in response to the sequence of ring indices and the sequence of moduli indices .

12. The method of claim 11 wherein the determining step further comprises the

steps of:

forming an integer equivalent I of the second block; and

iteratively dividing the integer into modulus coefficients.

60

13. The method of claim 11 further comprising the steps of:

segmenting said second block into a first and second portion, wherein said first portion is used to determine a sequence of moduli indices;

generating sign bits in response to said second portion; and

appending said sign bits to said moduli indices.

14. The method of claim 13 wherein said generating step comprises the step of differentially encoding said second portion.

15. The method of claim 11 further comprising the step of detecting a DC offset, and wherein the selecting step is performed in response to said DC offset.

16. The method of claim 11 wherein said signal point sequence has a length of six signal points.

17. An apparatus for mapping data bits to a sequence of signal points selected from a signal constellation wherein the points in the signal constellation are equally divided into rings, comprising:

a bit parser for segmenting input data bits into a first and second block;

a ring index generator for generating ring indices in response to said first block;

a multiple modulus converter for generating moduli indices in response to said second block, wherein at least one base modulus is other than a power of two;

signal point selection blocks for selecting points from the constellation in response to said ring indices and said moduli indices; and

a parallel to serial converter connected to said signal point selection blocks for

2╬╣ converting said points to a serial output.

18. The apparatus of claim 17 further comprising:

a sign mapper for appending a sign bit to said moduli indices;

and wherein said bit parser further segments input data bits into a third block, wherein said sign mapper generates said sign bits in response to said third block.

19. The apparatus of claim 18 further comprising:

a differential encoder connected between said bit parser and said sign mapper for differentially encoding said third block.

20. The apparatus of claim 18 further comprising:

an integrator connected to said parallel to serial converter for determining a DC offset.

21. The apparatus of claim 20 wherein said sign mapper is responsive to said integrator.

22. The apparatus of claim 20 further comprising:

a PCM to linear converter connected between said serial to parallel converter and said integrator to provide linear values to said integrator.

23. A method of mapping data bits to a sequence of signal points selected from a signal constellation wherein the points in the signal constellation are equally divided into rings, the method comprising the steps:

segmenting a plurality of data bits into a first block and a second block;

determining a sequence of ring indices in response to said first block using multiple modulus conversion, wherein at least one of said ring indices has a base modulus that is a value other than a power of two;

selecting a signal point sequence in response to the sequence of ring indices and said second block.

24. The method of claim 23 wherein the determining step further comprises the steps of:

forming an integer equivalent I of the second block; and

iteratively dividing the integer into modulus coefficients.

25. The method of claim 23 further comprising the steps of:

segmenting said second block into a first and second portion, wherein said first portion is used to determine a sequence of signal point indices;

generating sign bits in response to said second portion;

and wherein said selecting step further comprises selecting a signal point in response to said sign bits and said signal point indices.

26. The method of claim 25 wherein said generating step comprises the step of differentially encoding said second portion.

27. The method of claim 23 further comprising the step of detecting a DC offset, and wherein the selecting step is performed in response to said DC offset.

28. The method of claim 23 wherein said signal point sequence has a length of six signal points.

29. An apparatus for mapping data bits to a sequence of signal points selected from a signal constellation wherein the points in the signal constellation are equally divided into rings, comprising: y a bit parser for segmenting input data bits into a first and second block;

a signal point index generator for generating signal point indices in response to said first block;

a multiple modulus converter for generating ring indices in response to said second block, wherein at least one base modulus is other than a power of two;

signal point selection blocks for selecting points from the constellation in response to said ring indices and said signal point indices; and

a parallel to serial converter connected to said signal point selection blocks for converting said points to a serial output.

30. The apparatus of claim 29 further comprising:

a sign mapper for appending a sign bit to said signal point indices;

31. The apparatus of claim 30 further comprising:

32. The apparatus of claim 30 further comprising:

33. The apparatus of claim 32 wherein said sign mapper is responsive to said integrator.

34. The apparatus of claim 32 further comprising:

3╧è a PCM to linear converter connected between said serial to parallel converter and said integrator to provide linear values to said integrator

35. The modulus converter of claim 29 wherein said at least one modulus buffer is a random-access memory register.