**PROCESSING OF QUINARY DATA**

Curran, Philip (21 Kirkwood Sandymount Dublin 4, IE)

Molina Navarro, Alberto (Ateca, 15 Alcobendas E-MADRID, ES)

Mcelroy, Ciaran (12 Portersgate Avenue Clonsilla Dublin 15, IE)

Curran, Philip (21 Kirkwood Sandymount Dublin 4, IE)

Molina Navarro, Alberto (Ateca, 15 Alcobendas E-MADRID, ES)

*;*

**G06F7/48***;*

**H03H17/02***; (IPC1-7): H03H17/06; H03H17/02*

**H03H17/06**EP0277392A1 |

1. | A communication circuit comprising means for representing a PAM5 quinary data word in which a binary bit represents each of a word sign, requirement for a shift operation, and whether the word is zero. |

2. | A communication circuit as claimed in claim 1, wherein the circuit comprises a multiplier in which there is an AND gate for ANDing an input signal and the zerorepresenting bit. |

3. | A communication circuit as claimed in claim 1 or 2, wherein the multiplier further comprises an XOR gate for performing an XOR operation with the sign representing bit and an input signal to negate the coefficient if it is nonzero. |

4. | A communication circuit as claimed in claim 1 or 2, wherein the multiplier further comprises a multiplexer for leftshifting the coefficient using the shift representing bit and the output of the XOR gate. |

5. | A communication circuit as claimed in any of claims 2 to 4, wherein the multiplier further comprises means for correcting an LSB of the product where the data is negative. |

6. | A communication circuit as claimed in claim 5, wherein said correction means is in a sumofproducts block for a set of multipliers. |

7. | A communication circuit as claimed in any preceding claim, wherein the AND gate is connected to AND a multiplier and the zerorepresenting bit, the XOR gate is connected for performing an XOR operation with the signrepresenting bit and the output of the AND gate, and the multiplexer is connected for left shifting the coefficient using the shiftrepresenting bit and the output of the XOR gate. |

8. | A finite impulse response (FIR) filter comprising a tap delay line; a plurality of multipliers (1), wherein each multiplier comprises means for receiving a PAM5 quinary data word in which a binary bit represents each of a word sign, requirement for a shift operation, and whether the word is zero; and an SOP block, and in which each multiplier comprises : AND gates, each for ANDing an input coefficient and a zerorepresenting bit, XOR gates, each for performing an XOR operation with the signrepresenting bit and an input signal to negate the coefficient if it is zero, and a multiplexer for leftshifting the coefficient using the shiftrepresenting bit and the output of the XOR gate. |

9. | A finite impulse response (FIR) filter as claimed in claim 8, wherein the filter further comprises a correction circuit comprising means for maintaining a count of the sum of all sign values in the tap delay line, and means in the SOP block for adding said count to the sum. |

10. | A finite impulse response (FIR) filter as claimed in claim 9, wherein the correction circuit comprises means for maintaining said count by monitoring only the sign values of the first and the last tap delay line registers. |

Prior Art Discussion An example of where such processing is required is in a Finite Impulse Response (FIR) filter for echo or near end crosstalk (NEXT) cancellation in which a replica of a transmitted signal is generated. Fig. A represents, the generic structure of such a filter.

There are"taps", each having a delay line register datadel 1-datadel N through which symbols are shifted in series. A multiplier implements datadel X * coef Y, multiplying a coefficient coefl-coefN and the data. The multiplier outputs are summed to provide a sum-of-products (SOPs) filter output, providing a fresh sum value output for every clock cycle.

At a high level, implementation of such a filter is quite simple. However, in practice, several requirements and difficulties present themselves, such as the following.

Precision. An ASIC implementation must represent voltages as binary words of a certain length. The length of these words and the range which they cover will affect the precision of the filtering operation as well as the area (and hence power dissipation) of the filter.

Latency. If the output of the filter is needed immediately then the entire filtering operation is needed in one clock cycle (assuming the filter is clocked at the symbol rate).

Therefore all the logical operations need to be carried out in this clock cycle so a fast implementation will be required. Fast implementations normally involve some degree of parallelisation and hence consume more area and power. Normally however the output of the filter is not required immediately and the filter output can be generated over several clock cycles. In this case it is often possible to use a slower design with less parallelisation and hence the filter will consume less area and power.

Technology & Library. The range of logical gates available to the implementer is a function of the technology in which the ASIC is to be implemented and the cells (gate macros) available in the library for that process. For example certain cell macros exist such as full adders and half adders which are often more efficient in digital filters than logically equivalent blocks constructed from more basic logic blocks.

Layout. The place and route tools which are employed to position logical gates on the silicon substrate work more efficiently when applied to designs that replicate a simple structure many times over. This often leads to a design that occupies less area and consumes less power.

In a typical prior approach to implementation of a FIR for quinary data, as described in EP0911967 (T. I) two's complement is used for the multiplier. Such an approach provides a relatively high gate count, and there is consumption of relatively high power, in proportion to the gate count.

The invention is directed towards providing a circuit and method for processing the data to address the above requirements.

SUMMARY OF THE INVENTION

According to the invention, there is provided a communication circuit comprising means for representing a PAM-5 quinary data word in which a binary bit represents each of a word sign, requirement for a shift operation, and whether the word is zero.

In one embodiment, the circuit comprises a multiplier in which there is an AND gate for ANDing an input signal and the zero-representing bit.

In another embodiment, the multiplier further comprises an XOR gate for performing an XOR operation with the sign-representing bit and an input signal to negate the coefficient if it is non-zero.

In a further embodiment, the multiplier further comprises a multiplexer for left-shifting the coefficient using the shift-representing bit and the output of the XOR gate.

In one embodiment, the multiplier further comprises means for correcting an LSB of the product where the data is negative.

In another embodiment, said correction means is in a sum-of-products block for a set of multipliers.

In a further embodiment, the AND gate is connected to AND a multiplier and the zero- representing bit, the XOR gate is connected for performing an XOR operation with the sign-representing bit and the output of the AND gate, and the multiplexer is connected for left shifting the coefficient using the shift-representing bit and the output of the XOR gate.

According to another aspect, the invention provides a finite impulse response (FIR) filter comprising a tap delay line; a plurality of multipliers, wherein each multiplier comprises means for receiving a PAM-5 quinary data word in which a binary bit

represents each of a word sign, requirement for a shift operation, and whether the word is zero; and an SOP block, and in which each multiplier comprises :- AND gates, each for ANDing an input coefficient and a zero-representing bit, XOR gates, each for performing an XOR operation with the sign-representing bit and an input signal to negate the coefficient if it is zero, and a multiplexer for left-shifting the coefficient using the shift-representing bit and the output of the XOR gate.

In one embodiment, the filter further comprises a correction circuit comprising means for maintaining a count of the sum of all sign values in the tap delay line, and means in the SOP block for adding said count to the sum.

In another embodiment, the correction circuit comprises means for maintaining said count by monitoring only the sign values of the first and the last tap delay line registers.

DETAILED DESCRIPTION OF THE INVENTION Brief Description of the Drawings The invention will be more clearly understood from the following description of some embodiments thereof, given by way of example only with reference to the accompanying drawings in which:- Fig. 1 is a logic circuit diagram for a 5-bit multiplier; and Fig. 2 is a logic circuit diagram for a correction circuit.

In a FIR of the invention the quinary symbols are represented as set out in the table below. The representation is called Sign Shift Zero (SSZ) in this specification. Pam-5 Value Sign Shift Zero +2 O11 +1 001 0 000 -1 101 -2 111

The MSB indicates sign and is zero for positive numbers and one for negative numbers.

The middle bit represents a left shift operation which on 2C (twos compliment) numbers represents a multiplication by two. The LSB represents the presence of the number zero and is negated so that if the number is zero it too is zero, if the number is not zero then this bit is one. Table 1 below shows the output of the quinary multiplier for each of the five level inputs. Coefficient Quinary Value SSZ Output {c [4], c [3], c [2], c [1],c[0]} +2 011 {c[4],c [3], c [2], c [1], c [0], 0} {cM, c [3], c [2], c [1],c[0]} +1 001 {c[4],c[4], c [3], c [2], c [1], c [0]} {c [4], c [3], c [2], c [1], c [0]} 0000 {0, 0, 0, 0, 0, 0} {c [4], c [3], c [2], c[l],c [0]} -1 101 {!c[4],!c[4],!c[3],!c [2], ! c [1], ! c [O]} {c [4], c [3], c [2], c [l], C [01 1-2 Ill { ! c [4], ! c [3], ! c [2], ! c [1], ! c [0], 13 Table 1 A truth table for the quinary multiplier without LSB correction

In Table 1 where the quinary value is +2 the output of the multiplier is two times the input coefficients, and is shown in column 4. It can be seen that it is simply a left shift of the input coefficients with a zero appended. When the quinary value is +1 the output is equal to the input coefficients with the MSB repeated i. e. sign extended. For zero, the

output is zero. Where the quinary value is-1 the output is equal to the coefficient input bits inverted and the MSB is repeated. This is equal to one less than the arithmetic negative of the input coefficients. For a quinary value of-2 the output is the input coefficient bits inverted and left shifted and a 1 is appended.

This format is more efficient than 2C since each SSZ bit can be used directly in the filter multipliers.

Referring to Fig. 1 a 5-bit multiplier 1 comprises five AND gates 2, five XOR gates 3, and five multiplexers 4. Only the zero-bit line is ANDed with the coefficients, the sign- bit line is XORed, and the shift-bit line is multiplexed. As shown in Fig. 1, every bit of the coefficient can be ANDed directly with the zero bit of the data. The resultant will be zero if the zero bit is zero, or unchanged if it is not. The multiplexers 4 left shift the coefficient, and the sign bit is XORed with the coefficient to negate it. The arithmetic negative of a twos complement number can be obtained by inverting each of the bits and adding 1 to the LSBs.

Therefore in an SOP connected to the multiplier 1 an addition needs to be done where X LSBs need to be added to the result. X represents the number of negative data terms in the delay line. Heretofore, this has been performed by providing an adder associated with each multiplier of the FIR filter. Such an adder adds a the sign to the result of the every multiplication. Thus, for N taps there are N adders.

However, in the invention the sign bit S is used to achieve the same effect using only one correction circuit, shown in Fig. 2 and indicated by the numeral 10. The circuit 10 resides between the inputs to the tap delay line and the SOP block of the FIR filter. Its inputs are only the first and the last register valves data 0 (2 and data_39 (2). The output is a single word correction factor, delivered to the SOP block.

The data D (2 value is inputted to an XOR gate together with the data_39 (2) value, the latter being delayed by a register 11. The remainder of the circuit 10 effectively maintains a count of the sum of the all sign values in the tap delay line by incrementing and decrementing a counter 12 as the sign values in the first and last tap delay line registers change. The counter is incremented and decremented as follows, in which a sign value of 1 means negative and 0 means positive. data-0 (2) value data-39 (2) value Correction 0 0 no change 1 0 increment 0 1 decrement 1 1 no change Thus, in the circuit 10, if the sign values are the same a multiplexer 13 selects the previous count value, and if different, it selects the previous value together with the correction.

The size of the counter register should be chosen so that overflow will not occur. The reset signal is used to reset the counter to zero. This is a very simple circuit relative to the overall SOP circuitry.

It will be appreciated that the multiplier 1 of Fig. 1 is very simple and can be implemented with a very low gate count. It is thus a considerable saving on a multiplier using prior representation schemes such as magnitude or 2C representations. It will also be appreciated that the overall FIR filter is considerably simpler than in prior approaches.

The invention is not limited to the embodiments described but may be varied in construction and detail. For example, multipliers of the invention could be implemented for other suitable functions in a 1000BASE-T transceiver such as a base line wander compensation circuit, a decision feedback equaliser, or a decision feedback sequence estimator. Also, the order of the operations may be different from that illustrated in Fig.

1.

**Previous Patent:**FAST MEDIAN FILTER

**Next Patent: SYSTEM AND METHOD FOR REAL-TIME OBSERVATION ASSESSMENT**