DE10325835B4 - Method for modeling a model filter and circuit - Google Patents

Abstract

The method according to the invention has the following method steps: (a) A non-recursive model filter with a first number of coefficients, which simulates a filter chain connected downstream of the circuit for crest factor reduction, is provided; (b) The impulse response of the filter chain is determined; (c) From the impulse response determined in method step (b), filter coefficients for an optimized model filter are determined by approximation using a window of defined length in which the impulse response has the highest energy content. The invention also relates to a circuit for crest factor reduction which has such a model filter.

Description

The invention relates to a method for modeling an optimized model filter, which is arranged in a model path of a circuit for crest factor reduction of a multi-carrier data transmission system. The invention further relates to a circuit for crest factor reduction.

In modern telecommunications, the high-bit rate data transmission on a subscriber line plays an increasingly important role, in particular because it promises itself a larger usable bandwidth of the data to be transmitted combined with a bidirectional data communication.

One technology that has become increasingly important in recent times is the so-called multi-carrier data transmission, which is also known as "multi-carrier" transmission, as "Discrete Multitone (DMT)" transmission or as "Orthogonal Frequency Division Multiplexing (OFDM). "Transmission is known. Such a data transmission is used, for example, in line-connected systems, but also in the radio sector, for broadcast systems and for access to data networks. Such systems for the transmission of data with multi-carrier transmission use a multiplicity of carrier frequencies, wherein for the data transmission the data stream to be transmitted is split up into many parallel sub-streams which are transmitted in frequency multiplex independently of one another. These substreams are also referred to as Einzeltrager.

One representative of the multi-carrier data transmission is the ADSL technology, where ADSL stands for "Asymmetric Digital Subscriber Line". ADSL refers to a technique that allows the transmission of a high bit-rate bit stream from a central office to the subscriber and a bit rate bit stream passing from the subscriber to a central office. In this technique, the telecommunication line is divided into at least one channel for conventional telephone services (ie voice transmission) and at least one other channel for data transmission.

Although many problems have already been solved in such multi-carrier data transmission systems as ADSL, some problems still remain unresolved.

A problem associated with this multi-carrier data transmission results from the fact that as a result of the superimposition of very many individual carriers, these can briefly add up to very high peak values in the transmission signal. The peak-to-rms ratio is called the crest factor, and its square is called the PAR (Peak to Average Ratio). Especially with multi-carrier systems such as ADSL, the crest factor can become very large - for example greater than 6. Even though these peak values are very rarely present in the resulting amplitude and typically only for very short periods of time, they represent a major disadvantage of multicarrier data transmission.

A large crest factor causes various problems in the overall system of data transmission:
The maximum possible modulation of the digital / analog converters and of the analog circuit parts, for example filters and line drivers, must be designed in their modulation range and their dynamics or resolution for the maximum occurring peak values. This means that these circuit parts must be dimensioned much larger than the effective modulation. This is accompanied by a correspondingly high operating voltage, which also leads directly to a high power loss. Especially with line drivers, which generally have a non-negligible non-linearity, this leads to a distortion of the signal to be transmitted.

Another problem of data transmission at high crest factors is that a very high peak in the transmit signal may exceed the maximum possible modulation. In this case, a limitation of the transmission signal sets in - this is called a clipping. In these cases, however, the transmission signal no longer represents the original transmission signal sequence, so that transmission errors occur.

For this reason, in multi-carrier data transmission systems there is a need to suppress or avoid such peak values as much as possible. This problem is known in the literature under the term crest factor reduction or PAR reduction. There are several approaches to reducing the crest factor here:
In a known method, some carriers or carrier frequencies are reserved from the multi-carrier data transmission system (typically about 5% of the spectrum), which are then no longer available for data transmission. This means that these carriers are initially set to zero. From these reserved carriers a function in the time domain with the highest possible, temporally narrow peak value is generated, which forms the correction signal or the so-called kernel. Iteratively, this kernel, which occupies only the reserved carriers, is weighted with an amplitude factor proportional to the difference of maximum peak and desired maximum value and subtracted from the transmit signal in the time domain. In this case, the kernel is substituted for the corresponding peak value of the transmission signal used for the excessive crest factor is responsible, shifted cyclically. The shift rate of the DFT transformation ensures that only the reserved carriers are occupied even after the shift.

The method just described operates only in the time domain and is therefore characterized by its speed and low complexity. However, a significant disadvantage of this method is that subsequent circuit parts and filters for the determination of the kernel are no longer taken into account. These filters, which are always present in practice, in turn increase the crest factor of the signal superimposed on the kernel and thus cancel out part of the crest factor reduction. For this reason alone, the method described above is not suitable for practice.

In the international patent application WO 03/026240 A2 a method based on the method described above is described. By means of the correction signal, the high peak values in the time signal to be transmitted, which are responsible for a too high crest factor, are iteratively reduced. The advantage of the method described in WO 03/026240 A2 is that for the generation of the correction signal and thus for the crest factor reduction and subsequent filter functions are taken into account, which thus no longer to an excessive crest factor in after the overlay with the correction signal can lead to the time signal to be sent. For this purpose, a model filter is provided in the model path of the circuit for crest factor reduction, which simulates the downstream of the circuit for crest factor reduction filter chain.

In order to optimally emulate the filter chain downstream of the circuit for reducing the crest factor, a model filter of the highest possible order is typically used. Such a model filter has a multiplicity of filter coefficients. Here, recursive filters, for example an IIR filter (IIR = infinite impulse response), or non-recursive filters, for example a finite impulse response (FIR) filter, can be used as filters. Such filters can be formed in terms of hardware or implemented in a digital signal processor, since here the arithmetic units of the processors usually provide multipliers with downstream accumulators, with which the immediate execution of the discrete convolution is possible. In the case of a higher-order FIR filter, however, calculation of the correction signal requires a computational effort which, in the case of m coefficients, results essentially from multiplication to be performed m times. However, multiplications require more computational effort than other computational operations. On the one hand, a high computational effort reduces the performance of the entire system and, on the other hand, a correspondingly complex hardware is required for this, which leads directly to an increase in the chip area.

In the international patent application WO 01/82547 A1 a spread spectrum data transmission system is described. This data transmission system comprises a non-recursive model filter with a predetermined number of coefficients that mimic a downstream filter, wherein the impulse response of this downstream filter is determined. The data transmission system works "sample-based", ie based on samples. Furthermore, the data transmission system described there does not work completely without distortion, since a certain amplitude distortion of the transmitted signal is unavoidable.

Furthermore, in the German Offenlegungsschrift DE 102 55 687 A1 discloses another method of reducing the crest factor of a multi-carrier signal by calculating a correction quantity for the output signal using a maximum signal estimate.

Proceeding from this, the object of the present invention is to reduce the computational outlay in the model path, in particular in the model filter, of a circuit arrangement for crest factor reduction.

According to the invention, this object is achieved by a method having the features of patent claim 1 and a circuit having the features of patent claim 9.

The idea underlying the present invention is that for modeling the filter chain downstream of the circuit for reducing the crest factor, the latter need not necessarily be modeled identically. For a reduction of the computational effort, it has proven to be advantageous to emulate the filter chain in the signal path of the circuit arrangement for crest factor reduction by a suitable replacement filter as small as possible order. Very often, this filter chain is defined in the signal path up to a much higher than the oversampling used in the model path. It is therefore advisable to undertake an undersampling of the impulse response of the filter chain as an intermediate stage for a modeling of this filter chain and to use a replica with a filter model of sufficiently high order, but with a lower order than for the exact modeling of the filter chain in the signal path.

The present invention is based on the finding that in the case of the crest factor Reduction is crucial for the quality of the model filter to perform a faithful reproduction of the time domain characteristics, so the impulse response of the filter chain, and not the frequency response of the filter chain itself. It has also been shown that the components directly limiting the band in the filter chain have a dominant influence on the crest factor of the time signal to be transmitted. A reduction of the model filter to precisely these components, for example the high-pass components and at least the first stage of the low-pass filter, has virtually no influence on the shape of the impulse response of the filter and thus the quality of the crest factor reduction.

In the method according to the invention, therefore, first a window of defined length with the highest energy content is used as an approximation of corresponding length for the model filter from the sub-sampled impulse response of the filter chain. In a very advantageous embodiment, coefficients which are not taken into account, for example the first coefficients of the impulse response of the filter chain, are replaced by delay elements or even completely eliminated. The resulting time offset between model path and signal path does not or only insignificantly influences the performance of the method. In addition, the coefficients of the model filter thus obtained are quantized to a limited number of possible amplitude values, so that in the subsequent convolution operation, several samples can be added before multiplication by the filter coefficient. In this way, the number of multiplication per sampling step is significantly reduced. At the same computational effort, a much longer impulse response can be approximated.

Advantageously, a sequentially operating, non-recursive model filter, in particular a FIR filter, is provided, which processes the signal to be transmitted sequentially.

In an advantageous embodiment, the arithmetic unit is designed as a program-controlled unit, in particular as a microprocessor or microcontroller.

Advantageous embodiments and further developments are the dependent claims and the description with reference to the drawings.

The invention will be explained in more detail with reference to the exemplary embodiments indicated in the schematic figures of the drawing. It shows:

1 a block diagram of a circuit for crest factor reduction of a multi-carrier transmission system with inventive FIR model filter;

2 a canonical representation of an example of an adaptive quantization of an FIR model filter with 8 coefficients, which is reduced in its complexity by means of the inventive method to 3 multiplication steps;

3 a canonical representation of an example of an adaptive quantization of a realization of a polyphase FIR model filter with a 2-fold oversampling of the input signal;

4 a comparison of the impulse responses in the case of an FIR filter with 50 coefficients and in the case of a reduction according to the invention to 5 multiplications per polyphase with 2-fold oversampling of the input signal;

5 the quantization steps for the first polyphase component (a) and for the second polyphase component (b).

In all figures of the drawing, the same or functionally identical elements, signals and functions - unless otherwise indicated - have been designated the same.

1 shows a block diagram of a circuit for crest factor reduction - hereinafter referred to as CF circuit - a multi-carrier transmission system with inventive FIR model filter.

In 1 is with reference numerals 1 a section of a multi-carrier data transmission system called. With regard to the structure and operation of such a multi-carrier data transmission system 1 and in particular a CF circuit is to those already mentioned above WO 03/026240 A2 referred to, which is fully incorporated in the present patent application with respect to these objects.

In 1 is merely a section of the transmitter-side transmission path 5 shown, which is arranged between a transmitter, not shown, and a likewise not shown hybrid circuit which is connected to the corresponding telephone line.

In the transmission path 5 of the multi-carrier data transmission system are successively an IFFT module 2 , A circuit for crest factor reduction 3 and an output filter 4 arranged. The IFFT module 2 An input signal X0 supplied by the transmitter is supplied by the IFFT module 2 is modulated by inverse Fourier transform. The thus modulated input signal X, which has a sampling frequency of 2.208 MHz in the case of ADSL data transmission and a sampling frequency of 4.416 MHz in the case of ADSL + data transmission, becomes the downstream CF circuit 3 fed. The CF circuit 3 generates a crest factor-reduced output signal Z, which is the downstream output filter 4 is supplied. After filtering the crest factor reduced output signal Z, the output filter is output 4 a signal Z ', which is still Crest-factor reduced, although through this output filter 4 the output signal Z has been changed and thus the possibility of a renewed generation of an excessive

Crest Factors exists. This is only possible by a special design of the CF circuit 3 that just the action of this filter 4 taken into account by the output signal Z by the output filter 4 modeled as well as possible.

The CF circuit 3 has a model path for this purpose 6 on, which is parallel to a part 5 ' of the transmission path 5 is arranged.

The model path 6 branches at the input of the CF circuit 3 from the transmission path 5 ' off, leaving the model path 6 also the modulated input signal X is supplied. At the beginning of the model path 6 is a Uberabtastblock 11 provided that the original, formed by the user data time signal X L-fold, for example, 2-fold or 4-fold oversampled. Under certain circumstances, this Uberabtastblock can 11 also be waived.

The oversampled signal X 'becomes a downstream model filter 12 fed. In the model filter 12 it is a modeling of the output filter 4 or all, the CF circuit 3 downstream filter 4 or filter chains. The procedure for modeling this model filter 12 is described below on the basis of 2 to 5 described in detail.

The model filter 12 is an arithmetic unit 13 downstream, which searches from the oversampled and filtered signal the corresponding maximum point. For this purpose, a method is provided which essentially corresponds to that in the WO 03/026240 A2 as well as the method described above. Starting therefrom, correction signals Y _{CF are} generated, for example dirac-like pattern functions, which are stored, for example, in a memory. In the adding device 10 These correction signals Y _CF in the transmission path part 5 ' superimposed with the temporally suitably delayed transmission signal X with appropriate weighting. In this case, the correction signal Y _{CF is} shifted to the corresponding point of the time signal X, at which the maximum is currently located. The superimposition is then carried out, for example, by subtracting the correction signal Y _CF from the corresponding part of the time signal X in the adder 10 , The oversampled correction signal Y * _CF becomes in the adder 16 in the model path 6 superimposed with the temporally suitably delayed signal Y.

The feedback paths 17 . 18 in the send path part 5 ' and model path 6 as well as the appropriate switches 7 . 8th . 14 serve the iterative treatment of the respective signals. For storage or for buffering are the buffer devices 9 . 15 intended. The send path part shows 5 ' the iterative change of the non-oversampled time function X during the model path 6 performs the associated modification in the oversampled time slot X.

For the iterative treatment of the input signals X contains the transmission path part 5 ' a first switch 7 (Start), a second switch 8th (Stop), and one between these switches 7 . 8th arranged buffers 9 and adder 10 , The buffer device 9 serves the buffering, ie the delay of the input side supplied time signal X by a time delay in the model path 6 Account for the storage of the respective intermediate values from the iteration.

For the iterative treatment of the oversampled signals X 'is in the model path 6 a first switch 14 (Start), a buffer device 15 and an adder 16 intended. The computing and storage unit 13 leads just the correction signal and the adder 16 in which the correction signal Y * _{CF is} subtracted from the time signal Y in the model path. This results in a second iteration loop.

According to the invention, the model filter forms 12 the output filter 4 not exactly after, but provides a simplified model of the output filter 4 dar. The modeling of this model filter 12 is described below.

The following assumes that the model filter 12 a non-recursive filter, especially an FIR filter.

In a first approximation section, the actual impulse response of the output filter is determined 4 , which may well have recursive character, as already described above, cut out a window of length M, which has the greater energy content.

an die Koeffizienten des Ausgangsfilters 4 angepasst.Aim of further simplification of the model filter 12 is it, in a realization of a model filter 12 with M coefficients g = [g ₀ , g ₁ , ... g _M-1 ] ^T significantly reduce the number of multiplications required per filter step. For this, a set of K << M adaptive quantization levels is used L = {l ₀ , l ₁ , ..., l _K-1 } defined, which make up a proximity filter (model filter 12 ) with the filter coefficients G = [ G ₀ , G ₁ , ..., G _M-1 ] ^T with | G _l | Ε L can construct, which thus requires only k multiplications per filter step. The individual quantization stages I _k are in the following optimization step

to the coefficients of the output filter 4 customized.

The principle of this simplification is in the canonical representation in 2 graphed, where 2 (a) represents the filter in the initial situation (not optimized) and 2 B) denotes the optimized filter reduced in number of multiplications. As can be seen, the number of necessary additions does not decrease, or only when some of the coefficients are quantized to zero.

In the case of an oversampled input signal X is an efficient polyphase realization of the model filter 12 desirable as a practical implementation. Depending on the type of implementation, it may be necessary to apply the adaptive quantization separately to the individual polyphase filters g _m and to keep the number of quantization steps for the respective sets L _m constant in order to ensure a uniform utilization of the hardware. This is exemplary in 3 shown for a double Uberabtastung.

The form of the simplified FIR realization of the model filter in 3 has proven to be very efficient for crest factor reduction. First of all, the time domain properties of the model filter 12 and in particular the positions of the peak values are of particular importance, the coefficients can be very roughly quantized without significant performance loss. Advantageously, several coefficients can also be quantized to zero, which additionally reduces the computational effort.

In 4 is an example of the impulse response IA of a model filter 12 with 50 coefficients (n), in which, assuming a 2-times oversampled input signal X, the number of multiplications per polyphase is reduced from the original 25 to 5 (B). The 5 shows the assignment of the individual coefficients of the polyphase components to the respective quantization levels. 5 (a) shows the quantization _steps for the first polyphase _component I _0l (even coefficients) and 5 (b) shows the quantization steps for the second polyphase _{component 1111} (odd coefficients; 5 (b) ).

Although the present invention has been described above with reference to preferred embodiments, it is not limited thereto, but modifiable in many ways.

In particular, the invention is not limited to the above data transmission systems and methods but can be extended to all systems based on multi-carrier data transmission systems and methods for the purpose of crest factor reduction. In particular, the invention is not limited to an ADSL data transmission, but can be extended to all xDSL data transmissions. Also conceivable are mobile applications such as DAB (Digital Audio Broadcasting), DVB-T (= Digital Video Broadcasting-Terrestrial) or OFDM-based WLAN applications (Wireless Local Area Network).

Although the invention has been described above with reference to FIR filters, it is not limited thereto but can be extended to all possible non-recursive filters or transversal filters. Also, the numerical examples given for the number of filter coefficients and the number of reduced multiplication steps are only to be understood as examples and can of course be varied as desired within the scope of the invention. It is only essential that, in the case of the modeling method according to the invention, the number of multiplication steps carried out in the model filter is significantly reduced compared with the number of coefficients or the originally required multiplication steps.

Also, the invention is not limited exclusively to the specified high-pass filter, bandpass filter or low-pass filter, but can be extended to whatever form filter types.

Also, the invention is not necessarily limited to a 2-fold or 4-fold oversampling of the data signal to be transmitted. Rather, it can also be provided that no oversampling, even undersampling or an arbitrarily high oversampling takes place here.

LIST OF REFERENCE NUMBERS

1

More exchanger data transmission system

2

IFFT module

3

Circuit for crest factor reduction, CF circuit

4

Output filter, downstream filter chain

5

transmission path

5 '

Part of the transmission path

6

model path

7

switch

8th

switch

9

Buffering device, storage

10

adder

11

Block for oversampling

12

model filter

13

Computing and storage unit

14

switch

15

Buffering device, storage

16

adder

17

Iterationspfad

18

Iterationspfad

X0

input

X

(IFFT modulated) input signal

X '

(oversampled, modulated) input signal

Y

filtered signal

Z

output

Z '

filtered output signal

Y _CF

Correction signal for the transmission path

Y * _CF

oversampled correction signal for the model path

Claims (14)

Method for modeling an optimized model filter ( 12 ), which in a model path ( 6 ) a circuit for crest factor reduction ( 3 ) of a multi-carrier data transmission system ( 1 ), with the method steps: (a) a non-recursive model filter ( 12 ) having a first number (M) of coefficients which one of the crest factor reduction circuits ( 3 ) downstream filter chain ( 4 ) is provided; (b) The impulse response of the filter chain ( 4 ) is determined; (c) From the impulse response determined in method step (b), filter coefficients for an optimized model filter are determined by approximation of a window of defined length M in which the impulse response has the highest energy content. 12 ) certainly; characterized in that, after method step (c), the filter coefficients thus approximated are quantized to a second number (K) of possible amplitude values of the filter coefficients, the second number (K) being less than the first number (M) and the number of times optimized model filter ( 12 ) multiplications per filter step. A method according to claim 1, characterized in that only the band-limiting elements of the filter chain, in particular the high-pass component and / or the first stage of the low-pass filter, are considered for the determination of the impulse response in method step (b). Method according to one of the preceding claims, characterized in that for the quantization of the filter coefficients adaptive quantization levels are defined, from which the optimized model filter ( 12 ), which approximates the non-recursive model filter ( 12 ) and which has a second number (K) of multiplications per filtering step. Method according to one of the preceding claims, characterized in that in order to optimize the coefficients of the optimized model filter ( 12 ) each have a quantization step corresponding to the coefficients of the non-recursive model filter ( 12 ) is adjusted. Method according to one of the preceding claims, characterized in that filter coefficients of the filter chain ( 4 ), which are outside the window, are replaced by delays and / or set to zero. Method according to one of the preceding claims, characterized in that when an oversampling of the transmission signal (X) the optimized model filter ( 12 ) is implemented as a block of polyphase filters. Method according to one of the preceding claims, characterized in that the adaptive quantization are applied separately to the individual polyphase filters and the number of quantization steps per polyphase filter is kept the same compared to a filter without oversampling. Method according to one of the preceding claims, characterized in that the data symbol to be transmitted is a function of a plurality of signals provided within a given data frame and each of these signals is assigned to a carrier, each carrier occupying at least one frequency from a transmission data spectrum, wherein at least a carrier is reserved, which is not intended for data transmission. Circuit for crest factor reduction ( 3 ), in particular for carrying out a method according to any one of the preceding claims, the transmitting side in a multi-carrier data transmission system ( 1 ) is arranged (A) with a transmission path ( 5 ) with a signal (X) to be transmitted, (B) with a signal parallel to a section of the transmission path ( 5 ' ) model path ( 6 ), in which a non-recursive model filter ( 12 ), to which the signal (X) to be transmitted can be fed, which determines the characteristic of one of the circuits for crest factor reduction ( 3 ) downstream filter chain ( 4 ), which has a first number (M) of filter coefficients, the first number being less than the number of filter coefficients for an exact replica of the filter chain ( 4 ), the thus-approximated filter coefficients being quantized to a second number (K) of possible amplitude values of the filter coefficients, the second number (K) being less than the first number (M) and the number of the model filter being optimized ( 12 ) multiplications per filter step. with a model filter ( 12 ) downstream processing unit ( 13 ), one of the model filter ( 12 ) filtered signal (Y), which generates a correction signal (Y _CF ) dependent thereon, (C) with a subtracting device ( 10 ), the output side with the model path ( 6 ) and the transmission path ( 5 ) and which subtracts the correction signal (Y _CF ) therefrom to reduce the crest factor in the signal (X) to be transmitted. Circuit according to Claim 9, characterized in that the transmission path ( 5 ) and the model path ( 6 ) each have a feedback path ( 17 . 18 ) for the iterative formation of the correction signal (Y _CF ) for the transmission path ( 5 ) and the correction signal (Y * _CF ) for the model path ( 6 ) exhibit. Circuit according to one of Claims 9 or 10, characterized in that in the model path ( 6 ) one unity ( 11 ) is provided for oversampling, which before the model filter ( 12 ) is arranged and which performs an oversampling of the signal to be transmitted (X). Circuit according to one of claims 9-11, characterized in that delay elements are provided, which are not in the model filter ( 12 ) replace filter coefficients that are outside the window and before it. Circuit according to one of Claims 9-12, characterized in that a sequentially operating, non-recursive model filter ( 12 ), in particular a FIR filter, is provided which processes the signal (X) to be transmitted sequentially. Circuit according to one of claims 9-13, characterized in that the arithmetic unit ( 13 ) is designed as a program-controlled unit, in particular as a microprocessor or microcontroller.

Images