Precoding scheme for transmitting data using optimally-shaped constellations over intersymbol-interference channels

• US 5,388,124 A
• Filed: 06/12/1992
• Issued: 02/07/1995
• Est. Priority Date: 06/12/1992
• Status: Expired due to Term

• Current Assignee: Nariman Farvardin, Rajiv Laroia, RPX Clearinghouse LLC (RPX Corporation), Steven Alan Tretter
• Original Assignee: University of Maryland
• Inventors: Laroia, Rajiv, Tretter, Steven A., Farvardin, Nariman
• Primary Examiner(s): Chin, Stephen
• Assistant Examiner(s): Kim, Kevin
• Application Number:
  US07/897,477
• Time in Patent Office: 970 Days
• Field of Search: 375/37, 375/39, 375/34, 375/17, 375/11, 375/14, 379/99, 379/101, 379/60, 379/17
• US Class Current: 375/286
• CPC Class Codes: H04L 25/03343 Arrangements at the transmitter end
• CPC Class Codes: H04L 25/03993 Noise whitening
• CPC Class Codes: H04L 25/4975 Correlative coding using Tomlinson precoding, Harashima precoding, Trellis precoding or GPRS
• CPC Class Codes: H04L 27/3433 using an underlying square constellation

Abstract

A precoding scheme for noise whitening on ISI channels is presented where the scheme permits any type of shaping and is compatible with trellis coding. Its implementation complexity is significantly less than that of trellis precoding. The resulting reduction in shaping gain is small and becomes negligible as the rate per 2D of the constellation increases. This technique uses a structured vector quantization (SVQ) technique for optimal shaping of the constellation. Optimal (N-sphere) SVQ shaping results in higher shaping gains than those of Voronoi constellations based on known N-dimensional lattices. For a given CER₂ (or PAR₂) SVQ shaping results in optimal shaping. This is useful for implementation over QAM modems as significant shaping gains can be achieved even for a small constellation expansion. The SVQ-shaped constellations have a very reasonable implementation complexity. They allow for the possibility of opportunistic secondary channels, and there constituent 2D constellations are invariant to π/2 phase rotations (assuming cubic lattice based constellations). Compatibility of SVQ shaping with trellis coding was also demonstrated, resulting in the SVQ-shaped trellis-coded modulation scheme which can realize both shaping and coding gains. For transmission over ISI channels, SVQ shaping can be used together with the precoding scheme to realize shaping gains up to 1.15 dB.

Claims

1. A digital communication apparatus that transmits binary data from a transmitter to a receiver over a channel, the apparatus comprising:

means for receiving a stream of input binary data,

a shaping encoder means for mapping a block of data from the stream of input binary data to a point in a constellation of dimension N having a constituent d-dimensional constellation, the constellation point being represented as a concatenation of i constituent points each from the constituent d-dimensional constellation where i and d are integers;

wherein

i) boundary of the constituent d-dimensional constellation is chosen to minimize the peak energy for a given predetermined data transmission rate and a given predetermined d-dimensional constellation expansion ratio, and boundary of the constellation approximating a boundary that maximizes shaping gain;

ii) the mapping from the block of data to the constellation point is invertible and is performed by;

(a) partitioning the constituent d-dimensional constellation into t groups of points where t is an integer, each group containing equal number of points;

(b) assigning an integer length to each point in a group, the length being a non-decreasing function of the average energy of all the points in that group;

(c) assigning a unique index between 1 and t to each of the t groups;

(d) assigning a total length to each constellation point, the total length being sum of the lengths of its i constituent points;

(e) ordering all the constellation points such that;

i) constellation point with smaller total length precedes constellation point with larger total length, and. ii) for two constellation points with same total length, order depends on indices of groups that their i constituent points belong to,

(f) assigning a unique binary address to each constellation point depending on its position in set of ordered constellation points; and

(g) mapping the block of data to a constellation point that has same address as a binary number represented by the block of data.

2. The apparatus of claim 1 where the constellation points with same total length the ordering is further performed as a function of an i digit, base t number associated with each point, where i digits of the base t number are in a one-to-one correspondence with the i constituent points, each digit being one less than an index of a group that a corresponding constituent point belongs to.

3. The apparatus of claim 1 where N is a power of 2, d=2 and for constellation points with same total length, the ordering is further performed recursively by splitting each constellation point into two new points, each of half the dimension and mapping each new point to an address in a constellation of half the dimension;

the order of the constellation points being a function of two addresses corresponding to each constellation point.

4. The apparatus of claim 1 where d=2 and further all the points in each group that the constituent 2-dimensional constellation is partitioned into are points contained in a region approximating a shape of an annulus.

5. A digital communication apparatus that transmits binary data from a transmitter to a receiver over a channel using a signal set based on a predetermined trellis code, where the trellis code is generated by partitioning a d-dimensional constellation A₀ into two subsets B₀ and B₁ in the first level of set partitioning, the apparatus comprising:

i) means for receiving a stream of input binary data;

ii) a shaping encoder means for mapping a block of data from the stream of input binary data to a point in an N-dimensional constellation having B₀ as constituent d-dimensional constellation, the constellation point being represented as a concatenation of i constituent points each from the constituent d-dimensional constellation where N, i and d are integers;

wherein;

A) boundary or the constituent d-dimensional constellation is chosen to minimize the peak energy for a given predetermined data transmission rate and a given predetermined d-dimensional constellation expansion ratio, and boundary of the constellation approximating a boundary that maximizes shaping gain;

B) the mapping from the block of data to the constellation point is invertible and is performed by;

(a) partitioning the constituent d-dimensional constellation into t groups of points where t is an integer, each group containing equal number of points;

(b) assigning an integer length to each point in a group, the length being a non-decreasing function of the average energy of all the points in that group;

(c) assigning a unique index between 1 and t to each of the t groups;

(d) assigning a total length to each constellation point, the total length being sum of the lengths of its i constituent points;

(e) ordering all the constellation points such that;

i) constellation point with smaller total length precedes constellation point with larger total length, and ii) for two constellation points with same total length, order depends on indices of groups that their i constituent points belong to,

(f) assigning a unique binary address to each constellation point depending on its position in set of ordered constellation points; and

(g) mapping the block of data to a constellation point that has same address as a binary number represented by the block of data;

iii) means for pairing each of the i constituent points of the N-dimensional constellation point to a point in B₁ of the same energy as the constituent point, thereby generating a sequence of i pairs; and

iv) means for transforming the sequence of pairs, one pair at a time, into a trellis code sequence by choosing one point from each pair, where a point from a current pair is chosen based on a current trellis state which in turn is generated by a previous trellis state and a previous chosen point.

6. The apparatus of claim 5 wherein the channel is an intersymbol interference channel and further comprising a precoder means for modifying each sample of the trellis code sequence one at a time such that an input signal to a Viterbi trellis decoder in the receiver is a valid trellis code sequence where the said modifying of a current sample of the trellis code sequence is performed by subtracting a small dither sample which is a function of previous modified samples and the intersymbol interference characteristics of the channel.

7. A digital communication apparatus that transmits binary data from a transmitter to a receiver over a channel using a constellation based on a predetermined k-dimensional trellis code and having a constituent k-dimensional constellation A₀, where A₀ is partitioned into two subsets B₀ and B₁ in a first level of set partitioning used to generate the predetermined trellis code, where k is an integer, the apparatus comprising:

i) means for receiving a stream of input binary data;

ii) means for mapping the input binary data into sequence of pairs of points where each pair of points contains one point from B₀ and one point from B₁ ; and

iii) means for transforming the sequence of pairs of points, one pair at time, into a trellis code sequence by choosing one point from each pair where a point from a current pair is chosen based on a current trellis state which in turn is generated by a previous trellis state and a previous chosen point.

8. The apparatus of claim 7 wherein both points in each pair of points in the sequence of pairs of points have same energy.

9. A digital communication apparatus that transmits binary data from a transmitter to a receiver over an intersymbol interference channel, the apparatus comprising:

i) means for receiving a stream of input binary data;

ii) means for mapping the input binary data into a trellis code sequence from a predetermined trellis code based on partitioning a lattice into a certain number of cosets; and

iii) a precoder means for modifying each sample of the trellis code sequence one at a time such that an input to a Viterbi trellis decoder in the receiver is a valid trellis code sequence, where the said modifying of a current sample of the trellis code sequence is performed by subtracting a small dither sample which is a function of previous modified samples and intersymbol interference characteristics of the channel.

10. The apparatus of claim 9 wherein each dither sample which is subtracted from a sample of the trellis code sequence further lies inside the Voronoi region of a lattice of which the cosets are translates.

11. A digital communication apparatus that transmits binary data from a transmitter to a receiver over intersymbol interference channels with data processing components which are serially connected in the following order comprising:

a) a means for entering input binary data to the apparatus,

b) a data shaping encoder means for generating multidimensional signal constellation points,

c) a trellis encoder means,

d) a Laroia-Tretter-Farvardin (LTF) precoder means for shaping and modifying trellis encoded multidimensional signal constellation points before transmission,

e) a transmit filter means,

f) a channel filter means,

g) a linear equalizer-sampler means,

h) a noise prediction-error filter means H(z) for i) whitening noise introduced into the channel, ii) reducing the variance of the noise in the channel and iii) in combination with the LTF precoder, generating output binary data that is always a sequence consistent with the trellis encoder means,

i) a Viterbi trellis decoder means for decoding the noise plus trellis sequence into a noiseless trellis sequence,

j) an intersymbol interference removing filter 1/H (z) and a quantizer encoder means for performing the inverse operation of that performed by the LTF precoder means, and

k) a shaping decoder means for outputting the binary data.

12. The apparatus of claim 11 wherein the data shaping encoder means and the shaping decoder means are a structured vector quantizer based shaping technique that provides shaping of the multidimensional constellation.

13. A method of transmitting and receiving data in a communication apparatus employing an encoder and decoder within the apparatus, the steps including:

a) taking input binary data at the encoder and transforming by a structured vector quantizer decoder means for a rate r bits/dimension N-dimensional signal constellation shaping,

b) using a structured vector quantizer based encoding means that converts constellation points to binary data sequences at the decoder within the apparatus, and

c) mapping at the encoder that assigns a unique Nr-bit binary number to a codevector which is determined from a codebook Q that consists of 2^Nr codevectors, the decoder uses an inverse step of the encoder that assigns the codevector to the Nr-binary codeword wherein Nr is an integer.

14. The method of claim 13 wherein the mapping step comprises:

a) determining the given codebook Q with the set of lengths L of integer lengths, a desired per sample rate r, and a threshold L of integer value that is determined by counting grid points in a manner that limits a constellation expansion ratio in two-dimension until there are 2^Nr points, and

b) determining a largest total length for which all the grid-points of that length are included in a collection of the grid-points.

15. The method of claim 13 wherein the shaping of multidimensional signal constellations include the step of combining the constellation shaping with trellis coded modulation.