A system and method for filter taps

Abstract

A method includes receiving an input signal (206) at a filter, where the filter includes a plurality of filter taps (209A - 209N), and where each of a first filter tap (209A) and a second filter tap has a weighting coefficient (234). The method further includes turning off the first filter tap (209A) based on the weighting coefficient (234) of the first filter tap (209A) being below a threshold and the weighting coefficient (234) of the second filter tap being below the threshold, where the second filter tap is adjacent to the first filter tap (209A).

Description

Background Art

[0001] Ethernet is a network protocol that controls how data is transmitted over a local area network (LAN). Gigabit Ethernet transmits Ethernet frames at a rate of one gigabit per second. Data is transmitted over shielded or unshielded twisted pair cables. An Ethernet receiver receives data from a transmitter. The receiver includes an analog front end followed by a digital processing section. Summary of the Invention

[0002] In some examples, a system includes a filter configured to receive an input signal. The filter has a first tap, a second tap, and a third tap, and each of the first tap, the second tap, and the third tap has a weighting coefficient and a data gate. The data gate is configured to close the first filter tap in response to the weighting coefficients of the first filter tap, the second filter tap, and the third filter tap all being below a threshold, and wherein the second filter tap and the third filter tap are both adjacent to the first filter tap.

[0003] In some examples, a method includes receiving an input signal at a filter, wherein the filter includes a plurality of filter taps, and wherein each of the first filter tap and the second filter tap has a weighting coefficient. The method further includes closing the first filter tap based on the weighting coefficient of the first filter tap being below a threshold and the weighting coefficient of the second filter tap being below a threshold, wherein the second filter tap is adjacent to the first filter tap.

[0004] In some examples, a method includes determining the mean square error (MSE) of a signal in an Ethernet receiver. The method includes comparing the MSE with a target MSE. The method further includes changing a threshold for closing a filter tap in response to the MSE being higher than or lower than the target MSE and an error value. The method includes setting a threshold for closing a filter tap in response to the MSE being higher than the target MSE minus the error value or lower than the target MSE plus the error value. Brief Description of the Drawings

[0005] Figure 1 is a block diagram of a receiver in an Ethernet system in various examples.

[0006] Figure 2 is a block diagram of an echo canceller in various examples.

[0007] Figure 3 is a graph of echo coefficients of 216 echo canceller taps in various examples.

[0008] Figure 4 is a graph of echo canceller taps and echo coefficients in various examples.

[0009] Figure 5It is a graph of echo canceller taps and echo coefficients in various examples.

[0010] Figure 6 It is a table of performance results in various examples.

[0011] Figure 7 It is a flowchart of a state machine that provides an SNR-based iterative turn-off method in various examples.

[0012] Figure 8 It is a graph of echo canceller taps and echo coefficients in various examples.

[0013] Figure 9 It is a table of performance results for turning off filter taps using an adjacent method in various examples.

[0014] Figure 10 It is a schematic diagram of a turn-off method based on channel length in various examples.

[0015] Figure 11 It is a flowchart of a method for reducing power consumption in an Ethernet receiver in various examples. Detailed Description

[0016] In some Ethernet applications where low power consumption is useful, such as automotive applications. High-speed Ethernet implementations can consume a large amount of power. One technique for reducing power is to share the hardware used for filter taps in the echo canceller in the digital part of the receiver. Another technique for reducing power is to turn off these filter taps when some filter tap values are below a threshold. However, these techniques may not be able to reduce the power consumption to a more useful level and may cause the performance to drop below an acceptable threshold.

[0017] In the examples herein, certain filter taps in the Ethernet receiver can be turned off to reduce power consumption. The unwanted filter taps are distinguished from the wanted filter taps, and the unwanted filter taps can be turned off. Instead of using an absolute threshold to turn off the filter taps, a filter tap can be turned off if the filter tap and a certain number of adjacent filter taps each have a filter tap coefficient below a predetermined threshold. In an example, the predetermined turn-off threshold can be adjusted based on a target signal-to-noise ratio (SNR). In another example, the target SNR can be based on the Ethernet channel length. By using some or all of these techniques, the power consumption can be reduced while still maintaining an acceptable performance of the system.

[0018] In various examples herein, the receiver 100 ( Figure 1 shown therein) can be implemented in an Ethernet system. Figure 1 Illustrates the Ethernet physical (PHY) layer receive path. The transmitter path is not shown in Figure 1is shown in. Receiver 100 includes an analog front end 102 and a digital signal processing section 104. In operation, an input signal is received at the receiver input 106 via a network conductor (such as a local area network cable, an Ethernet cable, or some other network connection or bus). The input signal passes through a high-pass filter 108 (such as an analog high-pass filter). A coarse automatic gain control (CAGC) 110 provides gain to the input signal so that the amplitude of the input signal is at a level appropriate for other components in the receiver 100. Additionally, appropriate gain (with little or no added noise) will increase the signal-to-noise ratio of the input signal. In another example embodiment, the HPF 108 may include a low-noise amplifier (not shown) and a high-pass filter. In this example, the CAGC 110 is in the digital signal processing section 104.

[0019] The high-pass filter 108 provides the filtered input signal to a programmable gain amplifier (PGA) 112. The PGA 112 provides additional amplification to the input signal. The input signal is then provided to an analog-to-digital converter (ADC) block, which in this example includes ADCs 114A and 114B (collectively referred to as ADC 114). In one example, the receiver 100 has a symbol rate of 750 megahertz (MHz). The ADCs 114A and 114B are interleaved so that each ADC 114 operates at 375 MHz to provide a total throughput of 750 MHz. The clock signal in this example is provided by a 750 MHz phase-locked loop (PLL) 116. The clock signal passes through a phase interpolator 118, which may lag or lead the clock signal during operation. In one example, the phase interpolator 118 has a 6-bit control, which means it can adjust the phase of the clock signal by 2 6 = 64 different values. The clock recovery loop 120 in the digital signal processing section 104 sends phase up / down signals to control the phase interpolator 118.

[0020] The clock output from the phase interpolator 118 is divided at a clock divider 122 into a 375 MHz recovered clock signal. The 375 MHz clock signal is provided to the ADC 114. A timing recovery loop including an incremental ADC (iADC) delay control 124 and an iADC timing loop 126 (in the digital signal processing section 104) provides timing recovery functionality for the ADCs 114A and 114B.

[0021] The output signals from ADCs 114A and 114B are provided to FIFO (First In First Out) 128 in the digital signal processing section 104. FIFO 128 receives the signals from ADCs 114A and 114B and provides multiple paths (six paths for the purposes of this description) at its output. Only a subset of the six paths is shown here. Each of the six paths operates at 125 MHz (e.g., determined by dividing 750 MHz by the number of paths (6 in this example)), which provides a total operating speed of 750 MHz. Each of the six paths includes a gain component 130 (collectively referred to as gain components 130). For simplicity, only three gain components 130A, 130B, and 130C are shown. The gain of the gain components 130 can be adjusted with an iADC gain control signal provided by the iADC gain loop 132. The output signals of the gain components 130 are provided to the echo canceller 134.

[0022] Echo refers to the interference between the data transmitted and received on a channel. Echo may occur if the proximal transmit signal reflects from the transmit path to the receive path. Echo may also occur if at least a portion of the transmit signal on a single pair of twisted wires reflects back from the target device. The echo canceller 134 provides echo cancellation and is described in the following example. In some examples, echo cancellation may also be performed in the analog front end 102 ( Figure 1 not shown). The echo canceller 134 generates an echo estimate and attempts to cancel the echo by subtracting the echo estimate from the signal. An adaptive filter, such as a finite impulse response filter, can be used to model the response of the echo path.

[0023] After the echo canceller 134 provides echo cancellation, the echo canceller 134 provides signals on the six paths. In this example, six paths are used, but in other examples, there may be a different number of paths, or only one path. The six signals are provided to the fine automatic gain control (FAGC) components 136. For simplicity, two FAGC 136A and 136B (collectively referred to as FAGC 136) are shown instead of all six used in this example. The FAGC 136 provides fine gain control for the signals in conjunction with the gain loop 138. The six signals are then provided to the digital equalizer (DEQ) 140. The DEQ 140 equalizes the signals and then provides these signals to the feed forward equalizer (FFE) 142, which also equalizes the signals. In one example, the FFE 142 can use delay components to provide weighted signals that are fed forward and summed to provide equalization.

[0024] At the output of the FFE 142, a parallel decision feedback equalizer (DFE) 144 receives the output signal and stores the symbols in a shift register within the DFE 144. The DFE 144 uses a feedback loop with adders 146A and 146B (collectively referred to as adders 146; only two adders 146 are shown for simplicity) to remove inter-symbol interference (ISI) from the output signal. ISI occurs if one symbol interferes with another. After the DFE 144 removes the ISI, the symbols are sampled by limiters 148A and 148B (collectively referred to as limiters 148; two limiters 148 are shown, although there may be more limiters 148 in other examples), and then the symbols are provided to the next processing block of the Ethernet implementation.

[0025] Most of the current in the receiver 100 is consumed by the echo canceller 134, the FFE 142, and the DFE 144. Reducing the current in these components can help reduce the total current consumed by the receiver 100. In one example, the digital signal processing section 104 consumes approximately 250 milliamperes (mA) of current. The DFE 144 consumes approximately 80 mA of current, the echo canceller 134 consumes approximately 60 mA of current, and the FFE 142 consumes approximately 10 mA of current. In this example, each of these components includes a filter with a plurality of filter taps. The DFE 144 has 42 filter taps, the echo canceller 134 has 216 filter taps, and the FFE 142 has 4 filter taps. In other examples, these components may have different numbers of filter taps. According to the examples described below, if adjacent filter taps are below a certain threshold, the current consumption in the receiver 100 can be reduced by turning off the filter taps. In the examples herein, the DFE 144, the echo canceller 134, and / or the FFE 142 are configured to turn off filter taps in response to adjacent filter taps being below a certain threshold, thereby reducing power consumption. Now, these techniques will be described for Figure 2 these techniques.

[0026] Figure 2 is the echo canceller 200 in various examples. The echo canceller 200 is an example implementation of the echo canceller 134 from Figure 1 In other examples, the echo canceller 200 may include Figure 2Other components not shown in the figure. The echo canceller 200 includes a data path 202 and a least mean square (LMS) block 204. The transmitted data 206 enters a shift register 208 in the echo canceller 200. The transmitted data 206 is a transmitted signal that interferes with the signal on the receiving side (e.g., the receiver 100). In one example, the transmitted data 206 represents the original transmitted signal that reappears with a certain delay in the received signal (e.g., the echo). Once the echo is identified, it can be eliminated by subtracting it from the received signal. A digital signal processor, analog circuit, or software ( Figure 2 not shown) can identify the echo signal and convert it into a digital signal, such as the transmitted data 206. The echo canceller 200 is configured to cancel the interfering transmitted data 206. In this example, the transmitted data 206 is received in the form of symbols 0, 1, or -1, and these symbols are stored in the shift register 208. In this example, the shift register 208 has 216 filter taps 209, making it a 216-tap filter. For simplicity, Figure 2 four filter taps 209A to 209N are shown. The output of each filter tap 209 is multiplied by a weighting coefficient (hereinafter referred to as a coefficient) using multipliers 210A to 210N (collectively referred to as multipliers 210). The products of the multipliers 210 are added by an adder 212. The digital input signal 214 is added to the output of the adder 212 by an adder 216. The digital input signal 214 is an input signal from an ADC such as ADC 114 ( Figure 1 ). The digital input signal 214 includes an unwanted echo signal. The digital input signal 214 and the adder 216 can work to cancel the recovered echo signal. The output of the adder 216 is provided to an equalizer 218, and the equalizer 218 equalizes the signal. The output of the equalizer 218 is provided to a limiter 220, and the limiter 220 generates a limiter output 222.

[0027] The LMS block 204 is used to find the coefficients of the echo canceller 200 on the right side of the echo canceller 200. The limiter error of the limiter 220 is found using an adder 224. The limiter error is provided to the LMS block 204. For a given filter tap i209, the limiter error is multiplied by the transmitted data 206 from the filter tap 209 in the shift register 208 using a multiplier 226. The result of the multiplier 226 is scaled by an amplifier 228 and accumulated using an adder 230. The accumulated result is stored in a register 232. The register 232 contains the echo canceller coefficients estimated during training. Over time, the LMS block 204 generates coefficients 234 for a given filter tap i 209.

[0028] In an example, 216 LMS blocks 204 are used to determine 216 coefficients. To reduce power consumption, LMS sharing techniques are useful. Different from 216 individual LMS blocks 204, some hardware can share more than one coefficient in a time-division multiplexing manner. Thus, in the first clock cycle, the first LMS block 204 is used to update the first coefficient. In the second clock cycle, the same first LMS block 204 updates the second coefficient. The same first LMS block 204 is used to update the third coefficient in the third clock cycle and is used again to update the fourth coefficient in the fourth clock cycle. Thus, each LMS block 204 can be used four times, which is a technique called 4x LMS sharing. Using this technique, only 54 LMS blocks 204 are used instead of 216 LMS blocks 204. In another example, each LMS block 204 is used twice, which is called 2x LMS sharing. In some examples, 4x LMS sharing can save approximately 20 mA of current consumption.

[0029] The echo canceller 200 can also generate a mean squared error (MSE). The output of the adder 224 in the echo canceller 200 is the error of the limiter 220 and provides an example of the noise in the system. The MSE can be calculated from this noise. The output of the adder 224 is provided to the MSE block 236, which squares the error and then passes the output signal through a low-pass filter. The output of the low-pass filter is the MSE 238. The MSE 238 is a measure of the performance of the receiver 100 and can be used to adjust the threshold for turning off the taps 209, as described in the following examples.

[0030] In various examples, multiple techniques can be used to turn off the filter taps 209 to reduce current consumption. In one example, the data in the data path 202 can be cleared using zeros coupled to the taps 209 as shown. AND gates such as data gates 240A to 240N can be used to clear the filter taps 209 that are turned off to reduce current consumption. This technique is called data gating. Using data gating, zeros are passed to the multipliers 210A to 210N for the filter taps 209 that are data-gated, which produces zero outputs from the multipliers 210A to 210N. Those filter taps 209 are thus turned off by data gating.

[0031] Data gating can also be used in the LMS blocks 204 instead of the data path 202. The output signal from the adder 224 can be zeroed using the data gate 242 for a particular filter tap 209 that is to be turned off. Clearing the data in the LMS block 204 produces a zero coefficient 234, which turns off the filter tap 209.

[0032] In another example, clock gating can be used to turn off filter tap 209. The clock signal in the LMS block 204 provided to the register 232 can be turned off, which results in the coefficients output from the LMS block 204 being zero. Thus, either clock gating or data gating can be used to turn off the filter tap 209 to be turned off.

[0033] Figure 3 is a graph 300 of the echo coefficients of 216 echo canceller filter taps in various examples. The x-axis represents the number of filter taps in an echo canceller (e.g., echo canceller 134). The y-axis represents the value of the echo coefficient. The first curve 302 is a graph of a 6-meter Ethernet cable with two peaks. The first peak of curve 302 is near filter tap 20, while the second peak is near filter tap 70. The second curve 304 is a graph of a 15-meter Ethernet cable, which in this example is composed of a 6-meter cable and a 9-meter cable combined. Curve 304 has three peaks, the first peak is near filter tap 20, the second peak is near filter tap 70, and the third peak is near filter tap 140. The peaks in curves 302 and 304 are the peaks of the echo signals that the echo canceller 134 is configured to cancel. Thus, in the examples herein, the filter taps near the peaks remain on, so that the echo signals can be filtered and removed. The filter taps not close to the peaks are turned off to reduce current consumption.

[0034] The first peaks in curves 302 and 304 represent near-end echo. Near-end echo is the echo of the near-end transmitted signal at the receiver. In this example, the near-end echo appears near filter tap 20. The second peaks in curves 302 and 304 represent far-end echo. Far-end echo is the echo of the signal transmitted from the far end at the receiver. In this example, the far-end echo appears near filter tap 70. The far-end echo is delayed after the near-end echo, so it appears at a later filter tap number. The third peak only appears in curve 304. Curve 304 is a graph of a 15-meter Ethernet cable, which is composed of a 6-meter cable and a 9-meter cable combined. The third peak appears due to the reflection from the connection in the middle of the 15-meter combined cable.

[0035] As shown in graph 300, there are 216 filter taps, but only a small portion of the coefficients are significant. Thus, many of the coefficient values (on the y-axis) are zero or close to zero, as shown in graph 300. Therefore, dynamic shutdown of the insignificant filter taps can be performed to save power. For example, filter taps with coefficient values greater than 1 or less than -1 may be considered significant. In some systems, a blind threshold (e.g., ±1.0) is applied after the coefficients of the filter taps are found. For coefficients below the threshold, the filter taps associated with these coefficients are turned off in this type of system. One way to turn off a filter tap is to gate the clock or data of the filter tap as described above, and then the values of these filter taps become zero. Thus, only the significant filter taps remain on. This shutdown technique can be performed in both the echo canceller 134 and the DFE 144. However, a limitation of this solution is that if the filter taps are turned off using a blind threshold, the performance of the system degrades. The performance degradation is because some significant coefficients are still turned off because these coefficients are below the threshold (e.g., between 1 and -1). The blind threshold (e.g., the threshold found in some systems) causes performance degradation. If a filter tap with a significant coefficient is turned off, a portion of the echo may not be cancelled.

[0036] In the example herein, the decision to turn off a filter tap is determined by analyzing not only the filter tap under consideration but also adjacent (e.g., neighboring) filter taps. As Figure 2 shown, filter tap 209 receives the output data from shift register 208. In one example, the filter taps are arranged sequentially (209A, 209B, 209C,... 209N) in a row and receive the output data from shift register 208 according to their position in the row. Adjacent filter taps, neighboring filter taps, or filter taps adjacent to each other are filter taps that are consecutive in the row. For example, filter tap 209A is adjacent or neighboring to filter tap 209B. Filter tap 209B is adjacent or neighboring to both filter tap 209A and filter tap 209C. Filter tap 209C is adjacent or neighboring to both filter tap 209B and filter tap 209D. The filter taps within the two neighbors of filter tap 209C are filter taps 209A and 209B on one side and filter taps 209D and 209E on the other side. If the echo coefficient of an adjacent filter tap is above the threshold, the filter tap under consideration remains on. Similarly, if all of a predetermined number of adjacent filter taps are below the threshold, the filter tap is turned off. For example, if the adjacent filter taps (n + k to n - k) are below the threshold, the nth filter tap is turned off. The value of k can be determined experimentally. Using this technique, the filter taps near the peak coefficient value are not turned off, and only the unwanted filter taps are turned off.

[0037] In one example, the value of k is 2, and filter tap 50 is the filter tap under consideration. If the coefficient value of filter tap 50 is higher than the threshold, filter tap 50 remains on. However, if the coefficient value of filter tap 50 is lower than the threshold, adjacent filter taps are considered to determine whether filter tap 50 should be turned off. When k is 2, filter taps 48, 49, 51, and 52 are considered. Filter tap 50 is turned off only when the coefficient values of each of filter taps 48 to 52 are lower than the threshold. If any coefficient value is higher than the threshold, filter tap 50 remains on. If all the coefficient values of filter taps 48 to 52 are lower than the threshold, filter tap 50 is turned off. This analysis is performed for each of the 216 filter taps. Instead of using a blind threshold to turn off, filter taps close to those with significant coefficient values are kept on. In this case, the definition of close is determined by the value of k. A larger value of k will result in more filter taps being kept on compared to a smaller value of k. As described below, the examples in this document result in lower current consumption while maintaining performance above an acceptable threshold.

[0038] For curve 302, the filter taps located near filter taps 20 to 30 can be kept on because those coefficient values are significant or close to other significant coefficient values. Similarly, according to the examples in this document, the filter taps located near filter taps 65 to 75 can be kept on. For the 15 - meter cable represented by curve 304, the filter taps located near filter taps 20 to 30 and near filter taps 65 to 75 can be kept on. Additionally, for the 15 - meter cable, the filter taps near filter taps 140 to 155 can also be kept on.

[0039] Figure 4 It is a graph 400 of echo canceller filter taps and echo coefficients in various examples. The x - axis represents the number of filter taps in an echo canceller (such as echo canceller 134). The y - axis represents the value of the echo coefficient. The first curve 402 indicates the coefficients of the filter taps without turning off any filter taps.

[0040] The second curve 404 indicates a system using blind dynamic turn - off. For this system, filter taps with coefficient values lower than a certain threshold are turned off without considering any adjacent filter taps. On curve 404, the filter taps between approximately 30 and 40 are shown with a coefficient value of zero, meaning these filter taps are turned off using the blind dynamic turn - off of this system.

[0041] The third curve 406 indicates the dynamic turn-off of filter taps using adjacent filter taps according to the examples herein. By curve 406, a filter tap is turned off only if the adjacent filter tap is also below the turn-off threshold. As shown by curve 406, the filter taps between approximately 30 and 40 remain on, while those filter taps are turned off using the blind dynamic turn-off represented by curve 404. Similarly, some filter taps between approximately 145 and 170 are kept on using the method proposed herein, and those filter taps will be turned off using the blind dynamic turn-off represented by curve 404. Thus, in this example, turning off filter taps by considering adjacent filter taps results in more filter taps remaining on than the blind dynamic turn-off of a conventional system.

[0042] Figure 5 FIG. 500 is a graph of echo canceller filter taps and echo coefficients in various examples. The x-axis represents the number of filter taps in an echo canceller (e.g., echo canceller 134). The y-axis represents the value of the echo coefficient. Figure 5 The information in Figure 4 is the same as that conveyed in Figure 5 but

[0043] Figure 6 has a scaled-down y-axis to better illustrate the differences between the three curves. The first curve 502 indicates the coefficients of the filter taps without turning off the filter taps. The second curve 504 indicates a system using blind dynamic turn-off. According to the examples herein, the third curve 506 indicates the dynamic turn-off of filter taps using adjacent taps. As seen by curve 502, not turning off the filter taps does not turn on many filter taps, even if those filter taps have very small echo coefficient values. With the blind turn-off of curve 504, many filter taps are turned off even if they have significant echo coefficient values. As shown by curve 506, compared to curve 504, the filter taps remaining on more closely follow curve 502. Compared to curve 504, curve 506 keeps more filter taps on. However, many filter taps with unimportant values in curve 506 are turned off, and those filter taps are shown as having an echo coefficient value of zero in curve 506.

[0043] Figure 6Table 600 is a table of performance results in various examples compared to an alternative system. The blind close method will be compared to the adjacent method according to the examples herein. First, for a cable length of 6 meters, three echo coefficient thresholds are considered: no threshold (all filter taps remain on), threshold 1, and threshold 2. If all filter taps remain on, the SNR is 25.1 decibels (dB). If the echo coefficient threshold is set to 1, the blind close method results in 20 filter taps remaining on and the SNR is 24.9 dB. In the case of threshold 1, the adjacent method according to the examples herein results in 26 filter taps remaining on and the SNR is 24.8 dB.

[0044] If the echo coefficient threshold is set to 2, the blind close method results in 12 filter taps remaining on and the SNR is 22.9 dB. In the case of threshold 2, the adjacent method according to the examples herein results in 14 filter taps remaining on and the SNR is 24.5 dB. Using the adjacent method, two additional filter taps remain on in this example and the SNR is higher compared to the alternative method.

[0045] Second, for a cable length of 15 meters (9 - meter cable and 6 - meter cable), if all filter taps remain on, the SNR is 22.9 dB. If the echo coefficient threshold is set to 1, the blind close method results in 40 filter taps remaining on and the SNR is 22.3 dB. In the case of threshold 1, the adjacent method according to the examples herein results in 50 filter taps remaining on and the SNR is 22.8 dB.

[0046] If the echo coefficient threshold is set to 2 for the 15 - meter cable, the blind close method results in 20 filter taps remaining on and the SNR is 20.1 dB. In the case of threshold 2, the adjacent method according to the examples herein results in 26 filter taps remaining on and the SNR is 22.1 dB. In this example, using the adjacent method, six additional filter taps remain on and the SNR is higher compared to the alternative method.

[0047] Figure 7 State machine 700 provides an SNR - based iterative close method in various examples. In some examples, one threshold for the echo coefficient to be closed is not optimal for all cable lengths and may not be optimal for a cable formed by multiple cable segments. The SNR at receiver 100 varies with the cable length (e.g., the shorter the cable length, the higher the SNR).

[0048] In various examples herein, the threshold can be fine-tuned based on the target MSE. MSE is a measure of system performance. A high MSE indicates that the error in the system is too large, and more filter taps should be kept on to reduce the error in future iterations. A low MSE indicates that the error in the system is low, and some additional filter taps may be turned off to reduce current consumption. Turning off additional filter taps may increase the MSE, but if the MSE is below an acceptable level, a trade-off can be made to reduce current consumption until the MSE reaches the target level. The state machine 700 can incrementally increase or decrease the threshold for turning off echo coefficients. This threshold is adjusted until the target MSE is reached. At this point, the threshold is set, and the system operates in a steady state.

[0049] First, state 710 represents the idle state. State 720 is the training step. In state 730A, the dynamic off threshold is determined based on the target MSE. The threshold is the echo coefficient level that is compared with the echo coefficients of the filter taps to determine whether a filter tap is significant. MSE is the reciprocal of SNR (MSE = 1 / SNR). State 730B is an example of a method for determining the threshold. First, a target MSE is set, as well as a λ error value representing the error range within which the MSE is acceptable. During training, the MSE of the system is determined (as shown above). Figure 2 If the MSE is below the target MSE minus the λ error value, the threshold (Thr) is increased by a Δ value. In other words, the MSE is below the target, so the error in the system is better than expected in this example. Since the MSE is below the target, the threshold for the echo coefficients may be increased, so fewer filter taps remain on during dynamic turn-off (e.g., only the taps with echo coefficient values higher than the new threshold remain on). Fewer filter taps remaining active will result in a higher MSE, but as long as the MSE remains below the acceptable range, a trade-off is made for fewer filter taps to reduce current consumption while still remaining within acceptable performance parameters.

[0050] On the other hand, if the MSE is above the target MSE plus the λ error value, the threshold (Thr) is decreased by a Δ value. Decreasing the threshold will activate some currently inactive filter taps, which will reduce the MSE in the system. In one example, as described above regarding Figure 6As described above, using the adjacent method, a 6-meter cable with a threshold of 2 results in 14 filter taps being active. Reducing the threshold to 1 causes 26 filter taps to turn on. The additional 12 filter taps that are activated are filter taps that have an echo coefficient between 1 and 2 and satisfy the adjacent method described herein (e.g., adjacent values are also considered when determining whether to activate a filter tap). In this example, 12 additional taps are activated, which reduces the MSE in the system. Repeat the process in state 730B until the MSE is within an acceptable range determined by the λ error value. If the MSE is within the acceptable range, the state machine 700 advances to the stable state at state 740. In one example, a timeout can also be implemented. If state 730B does not produce an MSE within the acceptable range within a given amount of time, state 730B may time out and advance to state 740.

[0051] Moreover, the results may change during the stable state 740 and drop below the acceptable range. For example, it can be monitored and determined whether the MSE rises above a predetermined acceptable level (e.g., an MSE threshold). If this occurs, the state machine 700 advances to state 750. State 750 is a recovery state that occurs in response to the bit error rate (BER) or the MSE rising above the acceptable level. In the recovery state, all filter taps of the echo canceller 134 and the DFE 144 are turned on. Then, the state machine advances to state 730A to perform the training process again as described above.

[0052] In some examples herein, Figure 7 the training time of the method described may take 30 to 45 milliseconds. However, in certain applications, the link time is 100 milliseconds, so the increased training time still occurs within an acceptable time range.

[0053] Figure 8 is a graph 800 of the echo canceller filter taps and echo coefficients in various examples. The x-axis represents the number of filter taps in the echo canceller (e.g., echo canceller 134). The y-axis represents the value of the echo coefficient.

[0054] Curve 802 is a curve illustrating the adjacent filter tap turn-off method for a 15-meter cable formed by four cable segments as described herein. Curve 802 has peaks near filter tap 45, filter tap 90, and filter tap 120. Curve 804 is a curve illustrating the adjacent filter tap turn-off method for a 15-meter cable formed by two cable segments as described herein. Compared to the peaks in curve 802, the peaks of curve 804 are at different filter tap positions. For example, curve 804 has peaks near filter tap 20, filter tap 70, and filter tap 140.

[0055] Figure 9 Table 900 shows the performance results of closing filter taps using the adjacent method in various examples of this document. Various cable lengths are shown, with single cable lengths of 1 meter and 6 meters. Table 900 also shows a 15-meter cable length composed of two cable segments and four cable segments, and a 21-meter cable length composed of four cable segments. In this example, the target SNR is 18 dB. Therefore, an SNR of 18 dB or greater is considered an acceptable performance threshold.

[0056] In the case of a one-meter cable length, not closing the echo canceller filter taps provides an SNR of 26.2 dB, while closing the echo canceller filter taps using the method proposed in this document results in an SNR of 24.9 dB, with only 8 active filter taps. Therefore, for a one-meter cable, current consumption can be significantly reduced while still maintaining the SNR well above the target SNR of 18 dB. For a 6-meter cable, not closing provides an SNR of 25.3 dB. Closing the echo canceller filter taps using the method proposed in this document results in an SNR of 21.1 dB, with only 8 active filter taps for the 6-meter cable.

[0057] For the 15-meter cable composed of two cable segments, not closing provides an SNR of 22.9 dB. Closing the echo canceller filter taps using the method proposed in this document results in an SNR of 18.7 dB, with 12 active filter taps for the two-segment 15-meter cable.

[0058] For the 15-meter cable composed of four cable segments, not closing provides an SNR of 22.9 dB. Closing the echo canceller filter taps using the method proposed in this document results in an SNR of 18.2 dB, with 40 active filter taps for the four-segment 15-meter cable.

[0059] Finally, for the 21-meter cable composed of four cable segments, not closing provides an SNR of 20.3 dB. Closing the echo canceller filter taps using the method proposed in this document results in an SNR of 18.1 dB, with 80 active filter taps for the four-segment 21-meter cable.

[0060] Figure 10 Figure 1000 is a schematic diagram of the closing method based on channel length in various examples. In the above dynamic closing method, power is saved during steady-state operation. However, during training, all filter taps are turned on and consume power. If the cable length is known, some filter taps may also be turned off during training. For example, for a one-meter cable, some filter taps are not expected to be important because there is no far-end echo that must be cancelled (e.g., see above Figure 3(description). A 15-meter cable can use active filter taps to handle far-end echo, but these filter taps are not necessarily used for a 1-meter cable. Therefore, certain filter taps can be turned off during training according to the cable length to further reduce current consumption.

[0061] In one example, the receive gain (AGC) 1002 can be used to determine the filter tap length. In the schematic diagram 1000, the AGC 1002 provides a receive gain exponent 1004 to the look-up table 1006. Based on the receive gain exponent 1004, the look-up table 1006 provides a DFE tap length 1008 and an echo canceller tap length 1010. The DFE tap length 1008 and the echo canceller tap length 1010 indicate which filter taps can be turned off during training according to the cable length. For shorter cable lengths, more filter taps can be turned off.

[0062] Table 1050 is an example of receive gains for different cable lengths. For smaller cable lengths, the signal attenuates somewhat. For example, for a 1-meter cable, the signal attenuates by 9 dB. Therefore, the gain is an estimate of the channel length. The receiver gain is used in the look-up table to find the cable length and determine that filter taps beyond a certain tap value may be turned off according to the cable length. In one example, two look-up tables are used, one for the echo canceller filter taps and one for the DFE filter taps. According to the receiver gain, the subsequent filter taps can be turned off during training.

[0063] In another example, different cable types can have different look-up tables to determine which filter taps to turn off during training. For example, shielded cables and unshielded cables may have different insertion losses. Due to the different insertion losses, different amounts of gain may be useful for these different types of cables of the same length. However, regardless of whether the cable is shielded or unshielded, the same filter taps will remain on for a given length of cable. Therefore, if one type of cable has more receiver gain than another type of cable of the same length, the look-up table should reflect the same number of filter taps that should remain on for each cable type, even though the receiver gains are different.

[0064] Figure 11 is a flowchart of a method for reducing power consumption in an Ethernet receiver in various examples. The steps of method 1100 can be executed in any suitable order. In one example, the hardware components described above with respect to Figure 1 and Figure 2 can execute method 1100.

[0065] Method 1100 begins at 1110, where a filter receives an input signal, where the filter includes a plurality of filter taps, and where each of a first filter tap and a second filter tap has a weighting coefficient. As described above with respect to Figure 2 In one example, as described in the description of Figure 2 , the filter within echo canceller 134 receives an echo signal and filters the echo signal using filter taps with weighting coefficients.

[0066] Method 1100 continues at 1120, where the first filter tap is turned off based on the weighting coefficient of the first filter tap being below a threshold and the weighting coefficient of the second filter tap being below a threshold, where the second filter tap is located next to the first filter tap. As described above, turning off filter taps by considering whether adjacent filter taps are below a threshold produces acceptable results (e.g., acceptable SNR) while reducing current and power consumption.

[0067] The term "coupled" is used throughout the specification. The term can encompass a connection, communication, or signal path that implements a functional relationship consistent with this description. For example, if device A generates a signal to control device B to perform an action, then in a first example, device A is coupled to device B, or in a second example, if intermediate component C does not substantially change the functional relationship between device A and device B, then device A is coupled to device B through intermediate component C such that device B is controlled by the control signal generated by device A via device A.

[0068] A device "configured to" perform a task or function can be configured (e.g., programmed and / or hardwired) by a manufacturer at the time of manufacture to perform the function and / or can be configurable (or reconfigurable) by a user after manufacture to perform the function and / or other additional or alternative functions. The configuration can be by programming the device's firmware and / or software, by the construction and / or layout of hardware components, and the interconnection of the device, or a combination thereof.

[0069] As used herein, the terms "terminal", "node", "interconnection", and "pin" are used interchangeably. Unless specifically stated to the contrary, these terms are generally used to denote an interconnection or its terminals between device elements, circuit elements, integrated circuits, devices, or other electronic or semiconductor components.

[0070] Unless otherwise specified, "about", "approximately", or "substantially" before a value means + / - 10% of the stated value. Modifications to the described examples are possible within the scope of the claims, and other examples are possible.

Claims

1. A system for a filter tap, comprising: a filter configured to receive an input signal, the filter having a first filter tap, a second filter tap, and a third filter tap, and each of the first filter tap, the second filter tap, and the third filter tap having a weighting coefficient and a data gate; and wherein the data gate is configured to close the first filter tap in response to the weighting coefficients of the first filter tap, the second filter tap, and the third filter tap all being below a threshold, and wherein the second filter tap and the third filter tap are both adjacent to the first filter tap; wherein the data gate is configured to close the first filter tap in response to the weighting coefficients of a fourth filter tap and a fifth filter tap both being below the threshold, wherein the fourth filter tap is adjacent to the second filter tap, and the fifth filter tap is adjacent to the third filter tap.

2. The system according to claim 1, wherein the filter is within an echo canceller in an Ethernet receiver.

3. The system according to claim 1, wherein the filter is within a decision feedback equalizer in an Ethernet receiver.

4. The system according to claim 1, wherein the threshold is at least partially based on the signal-to-noise ratio of the Ethernet receiver.

5. The system according to claim 1, wherein the threshold is at least partially based on the channel length of the Ethernet system.

6. The system according to claim 1, wherein the data gate is configured to close the first filter tap by zeroing out the weighting coefficient of the first filter tap.

7. A method for a filter tap, comprising: receiving an input signal at a filter, wherein the filter includes a plurality of filter taps, and wherein each of a first filter tap and a second filter tap has a weighting coefficient; and closing the first filter tap based on the weighting coefficient of the first filter tap being below a threshold, the weighting coefficient of the second filter tap being below the threshold, and the weighting coefficient of a third filter tap adjacent to the first filter tap being below the threshold, wherein the second filter tap and the third filter tap are adjacent to the first filter tap.

8. The method according to claim 7, further comprising: closing the first filter tap by clock gating the weighting coefficient of the first filter tap.

9. The method according to claim 7, further comprising: closing the first filter tap by data gating the weighting coefficient of the first filter tap.

10. The method according to claim 7, wherein the threshold is at least partially based on the signal-to-noise ratio of the Ethernet receiver.

11. The method according to claim 7, wherein the threshold is at least partially based on the channel length of the Ethernet system.

12. The method according to claim 7, wherein the filter is within an echo canceller in an Ethernet receiver.

Images