On-line step-size calculation using signal power estimation and tone grouping of the frequency-domain equalizer for DMT-based transceiver

Abstract

An efficient method for calculating the step-sizes for a frequency-domain equalizer of a discrete-multitone communications system using signal power estimation and tone grouping (SPE-TG) while on-line. The SPE-TG method is used to calculate a plurality of subchannel step-sizes which are then stored in a lookup table. When on-line, the method uses signal power estimation to select step sizes for each tone, and uses these step sizes for frequency domain equalization. The SPE-TG method simplifies the calculations necessary for frequency domain equalization, thereby saving significant hardware and/or processing resources. The SPE-TG method is reliable and robust, and does not depend upon assumptions about the line, location, or channel.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to high-speed modem telecommunications. More specifically, the present invention discloses a method for efficiently calculating the step-sizes for a frequency domain equalizer in a discrete-multitone communications system.

2. Description of the Prior Art

Frequency domain equalization (FEQ) for discrete multi-tone (DMT) communications systems has involved separate processes for gain equalization (GE) and phase equalization (PE). In digital subscriber line (DSL) systems, FEQ must be performed when the communications are occurring, since each subscriber loop, typically twisted pair line, has individual characteristics for attenuation and frequency response due to bridge taps, outside interference, length of the line, and other factors. The frequency responses of several real-world ADSL loops are shown in chart form in FIG. 2. FEQ is performed during the training phase as the DSL modems on each end of the twisted pair line negotiate to determine the subchannel signal to noise ratio (SNR) and bit allocation.

Please refer to FIG. 7, which shows a block diagram of a typical DMT-based communications system. A time-domain equalizer TEQ shortens the channel dispersion to minimize intersymbol interference (ISI) caused by channel distortions such as loop length, gauge change, and bridge taps. Although TEQ can minimize ISI in the time domain, phase rotation and amplitude attenuation in the frequency domain still exist on each subchannel. To overcome these distortions, an FEQ is used to compensate for the phase and amplitude distortions on each subchannel.

However, existing methods for calculating FEQ are complex and use significant processing resources, either in software computing cycles or in specialized hardware such as high performance DSPs. One such prior art frequency domain equalization method is disclosed in U.S. Pat. No. 6,389,062 by Wu. The method uses two equalizers, a phase equalizer (PE) and a gain equalizer (GE), in order to implement frequency domain equalization. A block diagram of this prior art method is shown in FIG. 1. The FEQ input Y_k is processed by gain equalizer GE and phase equalizer PE.

In addition, the step-size used by the FEQ for training is a constant chosen through off-line simulation at the time of hardware design, with many uncertain issues left unresolved, and thus, the signal to noise ratio (SNR) under real-world conditions cannot be compensated for, resulting in larger net errors in phase and amplitude corrections, reducing real-world efficacy and reliability during negotiation.

Therefore there is need for improvement in frequency domain equalization for DMT-based communications systems to overcome the uncertainty of the prior art.

SUMMARY OF THE INVENTION

To achieve these and other advantages and in order to overcome the disadvantages of the prior art method in accordance with the purpose of the invention as embodied and broadly described herein, the present invention provides an improved method for on-line calculation of step size for DMT-based communications systems using signal power estimation and tone grouping (SPE-TG), the method being able to keep SNR losses below a user selectable limit of L-dB per tone, with L being a value chosen during design. The method predetermines a lookup table containing a set of step sizes to use for each tone, groups tones according to signal power estimation during the training period, and selects appropriate step sizes for each group, thereby keeping SNR losses to within a known, user-selectable, predetermined limit, which is compensated for with a system parameter, gain scaling, if the predetermined limit is chosen to be under 2.5 dB.

The present invention further provides for a reduction in computational complexity as compared to the prior art, using a single equalizer.

These and other objectives of the present invention will become obvious to those of ordinary skill in the art after reading the following detailed description of preferred embodiments. It is to be understood that both the foregoing general description and the following detailed description are exemplary, and are intended to provide further explanation of the invention as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention. In the drawings:

FIG. 1 is a block diagram of a prior art FEQ method;

FIG. 2 is a chart showing the frequency response of several sample ADSL channels under various real-world conditions;

FIG. 3 is a block diagram of the FEQ method of the present invention;

FIG. 4 is a chart showing the frequency response of the target channel of an example ADSL channel as analysed by the method of the present invention; and

FIG. 5 is a flowchart of an embodiment of the table generation method of the present invention;

FIG. 6 is a flowchart of an embodiment of the on-line method of the present invention;

FIG. 7 is a block diagram of a typical DMT-based communications system; and

FIG. 8 is a diagram illustrating the relationship of the method of the present invention to the training phases of signal negotiation between a pair of DMT-based modems.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

Reference will now be made in detail to the preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers are used in the drawings and the description to refer to the same or like parts.

According to Widrow et al., “Stationary and Nonstationary Learning Characteristics of the LMS Adaptive Filter,” Proceedings of the IEEE, vol. 64 no. 8, pp 1151-1162 (incorporated herein by reference), the bound of the step-size is inversely proportional to the total of tap signal input power, and must satisfy the criteria: $\begin{array}{cc}0<\mu <\frac{2}{\mathrm{total}\mathrm{of}\mathrm{tap}\mathrm{signal}\mathrm{input}\mathrm{power}}& \left(\mathrm{eq}1\right)\end{array}$

Essentially, the LMS algorithm is an instantaneous estimation to approximate the exact gradient for adaptation. The instantaneous gradient estimation leads to the gradient noise, which gives rise to the misadjustment. Therefore, considering the misadjustment effect, the criteria of (eq 1) can be reformulated as: $\begin{array}{cc}\mu \cong \frac{2M}{\mathrm{total}\mathrm{of}\mathrm{tap}\mathrm{signal}\mathrm{input}\mathrm{power}}& \left(\mathrm{eq}2\right)\end{array}$

where M is the amount of the misadjustment.

In addition, the number of iterations for convergence, τ_MSE, can be defined as $\begin{array}{cc}{\tau }_{\mathrm{MSE}}\cong \frac{1}{4·M}& \left(\mathrm{eq}3\right)\end{array}$

Consequently, a large step-size will lead to a large misadjustment but a short convergence time. Conversely, a small step-size will lead to a small misadjustment but a long convergence time. Thus, the choice of step-size involves a tradeoff between misadjustment and convergence time. From the perspective of managing the SNR, it is desirable to reduce the misadjustment, thus improving SNR, while also keeping convergence time at about the same level for each subchannel. However, channel characteristics differ from loop to loop and from subchannel to subchannel, depending on the individual characteristics of each subscriber loop (twisted pair line).

The SNR loss which results from the misadjustment, L, expressed in dB, can be determined by the system designer. If L is chosen to be less than 2.5-dB, the SNR loss can be compensated for using a system parameter, gk. The g_k parameter provides for adjustment from −2.5 to 2.5 dB, as defined in G.992.1 (ADSL), G.992.3 (ADSL2), G.992.5 (ADSL2+) and G.vdsl (VDSL). Therefore, the subchannel step-size can be calculated by: $\begin{array}{cc}{\mu }_k\cong \frac{\eta }{\mathrm{average}\mathrm{input}\mathrm{signal}\mathrm{power}\mathrm{on}k\mathrm{th}\mathrm{subchannel}}& \left(\mathrm{eq}4\right)\end{array}$

where eta, η, is 2M, and is obtained from the L parameter; for example, if L is chosen to be less than 0.25-dB, then η is equal to 0.118. Furthermore, the corresponding convergence time, per (eq 3), is then:τ_k≅5 iterations  (eq 5)

(eq 4) is of O(N) for the number of subchannels, i.e., it can be implemented in linear time with N_SC division operations for N_SC subchannels. However, (eq 4) implies that having equal average input signal power leads to having the same step-size. This in turn leads to the use of tone grouping to design the updating step-size based upon the average input signal power.

Please refer to FIG. 4. From the received signal power point of view, the target channel frequency response can be sliced horizontally into Q blocks to group the subchannels by average input signal power. Tones whose average input signal power levels are within the same block then use the same step size.

As described in the paper “On-Line Step-Size Calculation Using Signal Power Estimation—Tone Grouping for Frequency Domain Equalizer of DMT-Based Systems” by Wu et al. (the present inventors), 2004 IEEE Asia-Pacific Conference on Circuits and Systems, December 2004, and incorporated herein by reference, based on the SPE-TG approach, the subchannel step size as given in (eq 4) can be described by: $\begin{array}{cc}\begin{array}{c}{\mu }_k\cong \frac{\eta }{{Y_k}^2}\\ =\frac{\eta }{G_k^2\left[\left(2^b-1\right){\Gamma }_T\right]\widetilde{{\sigma }_k^2}},\end{array}& \left(\mathrm{eq}6\right)\end{array}$

where: Y_k and {tilde over (σ)}_k express the DFT output (FEQ input) and the standard deviation of the baseband equivalent channel noise on the k-th subchannel respectively; G_k represents the equivalent gain of the converged TEQ on the k-th subchannel; Γ_k denotes the total SNR gap including the SNR gap, system design margin and coding gain in power sense; and b is the value of the bit-allocation. Without losing the generality, G_k is considered the same for each subcarrier. Therefore, G=G_k and k=0, 1, . . . , N_SC−1. Also, practically, all of the subchannels have the equivalent AWGN. Consequently, {tilde over (σ)}={tilde over (σ)}_k and k=0, 1, . . . , N_SC−1.

Therefore the subchannel step-size can be further simplified to tone-grouping step-size, μ_i. The μ_i is given by: $\begin{array}{cc}{\mu }_i=\frac{\eta }{G^2\left[\left(2^b-1\right){\Gamma }_T\right]\widetilde{{\sigma }^2}},b\text{:}2\sim 16,i\text{:}1\sim 15& \left(\mathrm{eq}7\right)\end{array}$

Please refer to FIG. 3, a block diagram showing a representation of the method of the present invention. The FEQ utilizes the method of the present invention to update the step size for each subchannel in the updating hardware of the frequency domain equalizer.

Referring to FIG. 5, the method 100 for generating the lookup table comprises the following steps:

Step 110: for each subchannel from 0 to N_SC−1, calculate Γ_T as a function of the design parameters: SNR gap (Γ), design margin (γ_m) and coding gain (γ_c);

Step 120: for each subchannel from 0 to N_SC−1, calculate {tilde over (σ)} from BW, AWGN (−140 dBm/Hz) and receiver input impedance R (100-Ω);

Step 130: using the user-determined L parameter, calculate η

Step 140: assign the gain G;

Step 150: for each group number from 1 to 15, calculate the step size μ_i, where i is the group number 1 to 15.

Step 160: build a lookup table according to the results of Step 150.

In the method 100, b is not limited to less than 15 and its range is from 2 to 16 inclusive; furthermore G is typically kept within the practical range of about 0.5˜1.5.

Based on the method 100 above, an SPE-TG lookup table (LUT), described in Table 1, can be constructed with the assumptions of a 9.8-dB SNR gap, a 6-dB design margin, and a 0-dB coding gain. In addition, L is assumed to be less than 0.25-dB, resulting in η being equal to 0.118. From the VLSI implementation point of view, to simplify the necessary hardware, the step-size can be implemented as a power of two, listed in the fourth column of Table 1. Results of using the power of two approximations, shown in Table 2, are discussed below. Finally, there are 15 tone-groupings with power of two step-sizes ranging from 2⁶ to 2⁻⁸.

TABLE 1
Proposed SPE-TG Lookup Table.
		Floating	Power of 2
		Step-size	Step-size
	Signal Power	μflt, STE-TG	μpo2, STE-TG	Group
b	Pi, 1 ≦ i ≦ 15	μi, 1 ≦ i ≦ 15	Number
≦2	0.00100735	117.64279645	26	1
3	0.00235048	50.41834133	25	2
4	0.00503674	23.52855929	24	3
5	0.01040928	11.38478675	23	4
6	0.02115434	5.60203793	22	5
7	0.04264447	2.77896370	21	6
8	0.08562473	1.38403290	20	7
9	0.17158525	0.69066221	2−1	8
10	0.34350629	0.34499354	2−2	9
11	0.68734836	0.17241250	2−3	10
12	1.37503251	0.08618520	2−4	11
13	2.75040080	0.04308734	2−5	12
14	5.50113739	0.02154235	2−6	13
15	11.00261056	0.01077085	2−7	14
≧16	22.00555691	0.00538534	2−8	15

Referring to FIG. 6, a flowchart of an embodiment of the method of the present invention, the on-line method 200 comprises the following steps:

Step 210: for each subchannel from 0 to N_SC−1, calculate and accumulate the subchannel signal power, ${Y_k}^2$

Step 220: calculate the subchannel average signal power R_k

Step 230: assign a group number to each subchannel using a decision policy based on the power threshold values in column two of Table 1.

Step 240: for each subchannel from 0 to N_SC−1, select the step size μ_i for k-th subchannel according to the decision policy.

• • Step 250: perform FEQ using the step sizes determined in Step 240.

The decision policy of step 230 is described as follows:

i) R_k≧P₁₅, μ_k=μ₁₅

ii) P_i≦R_k≦P_i+1, μ_k=μ_i, for i=1, 2, . . . , 14

ii) R_k<P₁, μ_k=μ₁

where k and i are the index of the subchannel and group respectively; R_k and μ_k represent the average received signal power and step-size of the k-th subchannel respectively; and P_i and μ_i express the threshold value of the received signal power and step-size of the ith group respectively. The group step-sizes from μ₁ to μ₁₅ are stored in the LUT, which is obtained from method 100.

Please refer to FIG. 8, a diagram illustrating the relationship of the method of the present invention to the training phases of signal negotiation between a pair of DMT-based modems. The diagram illustrates an example using ADSL protocol. However, the present invention can be utilized with various protocols such as ADSL, ADSL2, ADSL2+, VDSL, or other type of protocol.

The initial protocol comprises a sequence of steps, including transceiver training. Transceiver training, in turn, comprises a sequence of steps defining exchange of known data so that line conditions can be determined. The relationship of the preferred embodiment of the method of the present invention is shown, with the step-size calculation of method 200 preferentially being performed during the beginning of the FEQ segment of the C-REVERB3 exchange during the transceiver training period. The results of the step-size calculation are subsequently used during the filtering and updating portions 410,420 of the FEQ training of the C-REVERB3 exchange and C-MEDLEY exchange respectively, as well as during any adaptation periods which may occur later.

The method of the present invention can be implemented either in software or in hardware. A lookup table can be used for storing the step sizes of Table 1 to speed calculations. In hardware, the method is realized with components for signal power accumulation, average signal power calculation, comparator (decision policy), look-up table (ROM table), filtering, updating, and error estimation. Power of two step-sizes can be substituted for the floating-point calculations to further simplify implementation in software or hardware, further reducing computational overhead with only a small difference in efficiency. Referring to Table 2, the resulting data rate difference between using the floating-point calculation and the integer calculation is less than ±0.21%.

TABLE 2
Data rate comparison between μflt and μpo2
	SPE-TG
	μflt	μflt, SPE-TG	μpo2, SPE-TG
Test Loop	REVERB/MEDLEY (kbps)	ε
T1.601 Loop #7	3096/3020	3076/3000	3096/3020	0.00%
T1.601 Loop #9	4204/4024	4184/4000	4212/4032	0.19%
T1.601 Loop	3896/3764	3900/3752	3916/3772	0.21%
#13
CSA Loop #4	9528/8620	9488/8640	9516/8616	−0.04%
CSA Loop #6	9048/8720	9016/8672	9052/8712	−0.09%
CSA Loop #7	8992/8692	8968/8672	8988/8696	0.04%
CSA Loop #8	8324/7840	8394/7796	8324/7832	−0.10%
mid-CSA Loop	12144/11304	12132/11288	12140/11304	0.00%

In comparison to the prior art, the method of the present invention uses only a single equalizer rather than a plurality of equalizers, thereby using significantly fewer computational resources and reduced hardware complexity. The present invention also compensates for actual subscriber loop conditions, as opposed to relying on the uncertainties of off-line simulations and field trials, which are unable to take into account the highly individual conditions of each real-world subscriber loop.

TABLE 3
Comparison between present invention and
method of U.S. Pat. No. 6,389,062
	U.S. Pat.
Item	No. 6,389,062	SPE-TG method
FEQ	GE plus PE	Single Equalizer
Architecture
Updating	GE	Direct-form	Amplitude and phase updating
	PE	LMS algorithm,	together with LMS algorithm
Computing	GE	2-real “mul”	8-real “mul”
Complexity	PE	8-real “mul”
Step-size	Single step-size	15 step-size groups predetermined
	predetermined by	by SPE-TG and calculated on-line
	off-line simulation	for each tone
Performance	uncertain issues which	SNRk loss less than 0.25-dB which
	depend on simulation	can be compensated for by gk
	model and field trial

The method of the present invention thus provides a significant improvement over prior art methods, resulting in faster convergence, simpler implementation, and improved adaptation to real-world subscriber loop conditions.

It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the scope or spirit of the invention. In view of the foregoing, it is intended that the present invention cover modifications and variations of this invention provided they fall within the scope of the invention and its equivalent.

Claims

1. A method for implementing a frequency-domain equalizer comprising: generating a plurality of step sizes; storing the plurality of step sizes in a lookup table; selecting a step size of a subchannel from the plurality of step sizes in the lookup table according to a received signal power while on-line; and applying the step size to the frequency-domain equalizer.

2. The method of claim 1 where generating the plurality of step sizes comprises: selecting a signal to noise ratio gap value; selecting a design margin value; selecting a coding gain value; calculating a baseband equivalent noise standard deviation value; selecting a signal to noise ratio loss value; and selecting a subchannel equivalent gain value.

3. The method of claim 1 where selecting the step size of the subchannel comprises: determining an eta value from the signal to noise ratio loss value; calculating a subchannel average input signal power value; and dividing the eta value by the subchannel average input signal power value.

4. The method of claim 1 where generating the plurality of step sizes comprises taking a square of the subchannel equivalent gain value multiplied by the total signal to noise ratio gap multiplied by a square of the baseband equivalent noise standard deviation value multiplied by a scaling factor value, where the scaling factor value is calculated by subtracting one from two raised to an integer power, where the integer is selected from the range [2 . . . 16] inclusive.

5. The method of claim 1 further comprising approximating the plurality of step sizes as powers of two.

6. The method of claim 1 where selecting the step size of the subchannel from the plurality of step sizes in the lookup table according to a received signal power while on-line comprises: calculating a signal power accumulation; calculating an average received signal power; comparing the average received signal power to a plurality of threshold values stored in a lookup table; when a threshold value of the plurality of threshold values is greater than the average received signal power value, selecting the subchannel step size associated with a chosen threshold value of the lookup table where the average received signal power is greater than or equal to the chosen threshold value and where the average received signal power is less than all threshold values in a set of a plurality of threshold values larger than the chosen threshold value; when each threshold value of the plurality of threshold values is less than the average received signal power value, selecting a maximum subchannel step size; and when each threshold value of the plurality of threshold values is greater than the average received signal power, selecting a minimum subchannel step size.

7. A method for implementing a frequency-domain equalizer comprising: generating a plurality of step sizes for a plurality of subchannels of a modem; storing the plurality of step sizes in a lookup table; and applying the plurality of step sizes to the frequency-domain equalizer.

8. The method of claim 7 further comprising: selecting a signal to noise ratio gap value; selecting a design margin value; selecting a signal to noise ratio loss value; and selecting a coding gain value.

9. The method of claim 7 where generating the plurality of step sizes of comprises: calculating a baseband equivalent noise standard deviation value; and calculating a subchannel equivalent gain value.

10. The method of claim 9 where calculating the step size of the subchannel comprises: determining an eta value from the signal to noise ratio loss value; calculating a subchannel average input signal power value; and dividing the eta value by the subchannel average input signal power value.

11. The method of claim 10 where generating the plurality of step sizes comprises taking a square of the subchannel equivalent gain value multiplied by the total signal to noise ratio gap multiplied by a square of the baseband equivalent noise standard deviation value multiplied by a scaling factor value, where the scaling factor value is calculated by subtracting one from two raised to an integer power, where the integer is selected from the range [2 . . . 16] inclusive.

12. The method of claim 7 further comprising approximating the plurality of step sizes as a plurality of powers of two.

13. The method of claim 7 where the plurality of step sizes are stored as integer powers of two in the lookup table.

14. The method of claim 7 where selecting the step size of the subchannel from the plurality of step sizes in the lookup table according to a received signal power while on-line comprises: calculating a signal power accumulation; calculating an average received signal power; comparing the average received signal power to a plurality of threshold values stored in a lookup table; when a threshold value of the plurality of threshold values is greater than the average received signal power value, selecting the subchannel step size associated with a chosen threshold value of the lookup table where the average received signal power is greater than or equal to the chosen threshold value and where the average received signal power is less than all threshold values in a set of a plurality of threshold values larger than the chosen threshold value; when each threshold value of the plurality of threshold values is less than the average received signal power value, selecting a maximum subchannel step size; and when each threshold value of the plurality of threshold values is greater than the average received signal power, selecting a minimum subchannel step size.

15. A method for implementing a frequency-domain equalizer comprising: generating a lookup table comprising a plurality of step sizes and a plurality of threshold values; calculating a signal power accumulation; calculating an average received signal power; and selecting a subchannel step size according to a decision policy.

16. The method of claim 15 where each step size of the plurality of step sizes is associated with one threshold value of the plurality of threshold values, and where each threshold value of the plurality of threshold values is associated with a single step size of the plurality of step sizes.

17. The method of claim 15 where the decision policy comprises: comparing the average received signal power to the plurality of threshold values; when a threshold value of the plurality of threshold values is greater than the average received signal power value, selecting the subchannel step size associated with a chosen threshold value of the lookup table where the average received signal power is greater than or equal to the chosen threshold value and where the average received signal power is less than all threshold values in a set of a plurality of threshold values larger than the chosen threshold value; when each threshold value of the plurality of threshold values is less than the average received signal power value, selecting a maximum subchannel step size; and when each threshold value of the plurality of threshold values is greater than the average received signal power, selecting a minimum subchannel step size.

18. The method of claim 15 where each step size of the plurality of step sizes is approximated by an integer power of two.