Improved blind SNR estimation
Background of the Invention
The invention relates to a method for estimating the signal to noise ratio (=SNR) ()of a modulated communication signal (rn) including a data symbol
component (sn) and a noise component (nn)
An overview of SNR estimation techniques is given in D R Pauluzzi, N C Beaulieu, IEEE Trans Comm Vol 48, Nr 10, P 1681-1691 (October 2000)
In order to provide optimal functioning of advanced mobile radio modules, accurate signal to noise ratio (=SNR) estimation of modulated communication signals is necessary On the physical layer, SNR values are used in maximum-ratio combining and turbo-decoding On higher layers, SNR values are used for call setup, macrodiversity and handover control
In environments with varying SNR, such as mobile radio networks, the determination of an SNR value must be rather quick
The SNR value  is defined as the ratio of signal power and noise power, i e
(Structure Removed)
SNR estimation can be done data-assisted (=DA) or Non-data assisted (NDA) The latter is often called a "blind" estimation
In the case of data assisted SNR estimation, a set of samples (with one sample typically corresponding to one bit) is known in advance After a transmission of the set of samples, the received data is compared with the original data by

means of a data-assisted maximum likelihood estimation The known set of samples can be a preamble or a training sequence
If the transmitted set of samples is not known in advance, a blind estimation algorithm must be applied Known algorithms of blind SNR estimation include
a) standard received data aided (=RDA) maximum likelihood SNR estimation (see e g D Pauluzzi, N Beauheu IEEE Trans Comm , Vol 48, No 10, pp 1681-1691, October 2000),
b) iterative SNR estimation (see eg B Li et al, IEEE Commun Lett Vol 6, No 11, pp 469-471, November 2002), and
c) Kurtosis SNR estimation (see e g R Matzner, K Letsch, Proc IEEE-IMS Workshop on Information Theory and Statistics, Alexandna/VA, USA, p 68ff, October 1994)
However, these known algorithms are rather inaccurate for low numbers of samples Moreover, in particular the iterative SNR estimation is cumbersome and time consuming
Summary of the Invention
It is the object of the present invention to provide a robust SNR estimation method for a modulated communication signal, wherein the method has a high accuracy even for low numbers of processed samples
This object is achieved by a method as introduced in the beginning, characterized in that an intermediate SNR value (RDA) of the modulated communication signal is derived from a data assisted maximum-likelihood estimation the assisting data not being known in advance but being reconstructed from samples of the modulated communication signal (rn),
and that an estimated SNR value (RDA_kR) is determined by a controlled nonlinear conversion of the intermediate SNR value (RDA)
The inventive method estimates the SNR of the modulated communication signal by means of a conversion of the intermediate SNR value The intermediate SNR value is obtained by a standard RDA maximum likelihood SNR estimation of the modulated communication signal The intermediate SNR value RD4 deviates from the true SNR value y In particular, for small values of the true SNR value , i e  → 0, RD4 is much larger than  This deviation is compensated for by the controlled non-linear conversion of the intermediate SNR value
The controlled non-linear conversion can be done by means of a conversion table based on an experimentally predetermined correlation between RDA and
 It is preferred, however, to model the correlation between RD4 and mathematically and to use the modelled correlation for conversion For modelling, information about and/or suitable assumptions for the characteristics of the modulated communication signal are useful In particular, the type of modulation should be known It has been found that signals modulated by binary phase shift keying (=BPSK) can be handled very well by the inventive method Moreover, the type of noise in the modulated communication signal can often be assumed to have a Gaussian distribution
With the aid of the highly accurate and robust estimated SNR values obtained over a broad SNR range, as determined by means of the invention, it is possible to improve receiver performance, which in turn allows for distressing of margins e g in network planning, in particular lower UL/DL transmission power requirements and larger distances between base stations, nodes B and the like
A highly preferred variant of the inventive method is characterized in that the controlled non-linear conversion is performed by a correction function , with RDA-ER =(RDA), wherein the correction function  is the inverse function of an estimated deviation function , with the estimated deviation function approximating a true deviation function ,true correlating the deviation of RD4
from , i e RD4 = true() and () * „u() Typically, the estimated
deviation function is determined by a mathematic model Its inverse function, i e the correction function, can be obtained by mirroring the estimated deviation function at the bisecting line of the first quadrant in the coordinate system If the estimated deviation function is simple enough, its inverse function can also be calculated analytically The estimated deviation function ()can be calculated by setting it equal to the intermediate SNR signal, which in turn is the ratio of the estimated signal power and the estimated noise power of the modulated communication signal The estimated signal power and estimated noise power, then, must be expressed as a function of y, with the latter
requiring suitable assumptions, such as e g an infinite number of samples to be processed
In a preferred further development of this method, T is chosen such that TO) = T^O) for large numbers of N, i e N -> <=o, with N being the number
of samples of the modulated communication signal (/,,) being processed The
estimated deviation function found in this way is then applied to data with a finite number of samples, too In many cases, in particular with a BPSK modulation and Gaussian noise distribution, true()can be determined
accurately for N → ∞ The latter assumption is justified in most real situations, when sufficiently large numbers of samples are available In particular, a number of 100 samples or more is sufficient
In a further preferred development of said variant of the inventive method,
(Formula Removed)
This choice of *¥(/) gives highly accurate
results in case of a BPSK modulation of the modulated communication signal
In an advantageous development of the variant, lF"' is applied by means of an approximation table An approximation table provides very quick access to values of the correction function, which are listed in the table Alternatively, online numerical or analytical calculation of values of the correction function is also possible, but more time consuming
Further preferred is a development of said variant wherein
(Formula Removed)
This hyperbolic function is a good approximation of true() in case of BPSK modulation Its inverse can be calculated straightforwardly as (Formula Removed) forf and (Formula Removed)
(Formula Removed) Thus, the correction function is available analytically, which simplifies and accelerates the determination of the estimated SNR value
Y RDA-FR
In another variant of the inventive method, the number N of samples of the modulated communication signal (r„) being processed is equal or less than
500, preferably equal or less than 100 In these cases, the inventive method already provides estimated SNR values of high accuracy, whereas known methods show worse accuracy For higher values of N, such as an N of 1000 or larger, it is worth mentioning that the inventive method provides equally
accurate estimated SNR values as known methods, but typically with less effort
Also in the scope of the invention is a computer program for estimating the signal to noise ratio() of a modulated communication signal (rn) according to
the inventive method The computer program may be saved on a storage medium, in particular a hard disk or a portable storage medium such as a compact disc
The invention also comprises a receiver system for estimating the signal to noise ratio (y) of a modulated communication signal (r„) according to the
inventive method The receiver system comprises a receiver unit The receiver unit can receive transmitted signals, with the transmission carried out by radio or an optical fiber line, e g The inventive method can be performed directly with the received transmitted signals, i e at the receiver unit Alternatively, the inventive method can be applied after a channel decoding, such as turbo decoding, of the received transmitted signals In the latter case, the method is performed with "soft" signals
Finally, the invention is also realized in an apparatus, in particular a base station or a mobile station, comprising an inventive computer program and/or an inventive receiver system as described above A typical mobile station is a mobile phone An inventive apparatus can be part of a 3G or B3G network, in particular a UMTS network or a WLAN network
Further advantages can be extracted from the description and the enclosed drawing The features mentioned above and below can be used in accordance with the invention either individually or collectively in any combination The embodiments mentioned are not to be understood as exhaustive enumeration but rather have exemplary character for the description of the invention
Brief Description of the Drawings
The invention is shown in the drawing
Fig 1 shows a binary transmission system with a noisy channel for use
with the inventive method,
Fig 2 shows plots of an estimated deviation function ()for a BPSK
real channel and a hyperbolic function ()HA() which approximates the former function, as well as their inverse functions in accordance with the invention,
Fig 3a shows a diagram plotting normalized mean square errors of
estimated SNR values with respect to the true SNR values as a function of the true SNR value, for standard RDA maximum likelihood SNR estimation (state of the art), inventive RDA-ER and inventive RDA-ERHA, with 100 samples processed per SNR estimation,
Fig 3b shows a diagram corresponding to Fig 3a, with 1000 samples per
SNR estimation,
Fig 4a shows a diagram plotting normalized mean square errors of
estimated SNR values with respect to the true SNR values as a function of the true SNR value, for standard RDA maximum-likelihood SNR estimation (state of the art), inventive RDA-ER, Iterative method (state of the art) and Kurtosis method (state of the art) with 100 samples processed per SNR estimation,
Fig 4b show a diagram corresponding to Fig 4a, with 1000 samples
processed per SNR estimation
Detailed Description of the Drawings
The invention deals with the estimation of SNR values in a transmission system, such as a radio telephone network A transmission system for use with the invention is shown schematically in Fig 1 At a source S, binary data is generated The binary data may contain information of a telephone call, for example The binary data consists of a number of bits bn, with n the index
number of the bits, running from 0 to vV -1, with N the total number of bits of the binary data Each bit may have a value of 0 or 1 In order to transport the binary data, it is modulated in a modulator M A typical modulation is the binary phase shift keying (BPSK) modulation, resulting in values of a data symbol component sn o +1 or-1 At the physical transmission of the data symbol
component sn, typically applying a carrier frequency, s„ is amplified by a factor
a Also, noise nn is superposed by the channel to the amplified data symbol
component α s„ The amplification factor a and the noise level are unknown
initially Thus in total, a modulated communication signal rn =α sn +nn is
generated and ready for detection at a receiver system The receiver system may be part of a mobile phone, for example During the physical transport, other data symbol components may be transmitted at the same time at other frequency ranges and/or at other (spreading) code ranges These other data symbol components can be neglected in this context, they may affect the noise component nn, though
In the following, a BPSK modulation is assumed as well as a real channel, with a noise probability density function (=PDF) of Gaussian type
(Formula Removed)
with a standard deviation or noise amplitude However, other signal characteristics are possible in accordance with the inverntion
When making a data-aided (DA) maximum-likelihood SNR estimation, an estimated amplitude a with known "pilots" s„, i e a set of N known data symbol component values, is calculated as
(Formula Removed)
with sn = {+1, -1} Note that a hat ^ above a value indicates an estimated value, and E is the estimation operation determining the mean value of its input values The estimated noise power a2 (second moment) is calculated as
(Formula Removed)
with v being a constant to be chosen according to literature (D R Pauluzzi, N C Beaulieu, I c ) as follows Maximum likelihood for σ2 estimation v = 0, Bias elimination for σ2 estimation v = 1, Minimum MSE for σ2 estimation v = -1, Bias elimination for y estimation v = 3, Minimum MSE for y estimation v - 5 The data-aided SNR estimation then results in
(Formula Removed)
In case of a "blind", i e received-data-aided (RDA) maximum likelihood SNR estimation the "pilots" are estimated based on receiver decisions, identical to the first absolute moment The estimated amplification factor α is calculated as
(Formula Removed)
Noise power is estimated correspondingly (compare DA above) Accordingly, an SNR value is estimated, with
(Formula Removed)
This value BPSKI RDA is the result of the standard RDA maximum likelihood SNR estimation, as known from literature
However, by using estimated data symbol component values sn instead of the true data symbol component values sn, error is introduced For sufficiently large SNR values , BPSK] RDA approximates y very well, but for small values of , the estimated values B PSKlRDA are too large
According to the invention, the BPSK1 RDA value is set equal to an estimated deviation function () For this purpose, the terms of BPSKI RDA are expressed as functions of the true SNR value y, requiring some approximations and assumptions
(Formula Removed)
In particular, the one-dimensional Gaussian type noise PDF as defined in the beginning has been introduced in this calculation, and it is based on the BPSK
modulation (indicated by the index BPSK1, where the "1" represents the one-dimensionality of the real channel and its noise)
The result  BPSK1 RDA of the standard RDA maximum likelihood estimation is used as a starting point for the actual calculation of an estimated SNR value BPSK1RDA-ER in accordance with the invention For this reason, the BPSK1 RDA is called an intermediate SNR value The "ER" index of BPSK\RD4-FR indicates an extended range, i e an improved range of application with the invention For calculating BPSKlRDi.tR, the inverse function of the estimated deviation
function () is determined, and the intermediate SNR value BPSKIRDA is assigned to 1, with
(Formula Removed)
Note that (x) = 0 for x The correlation of BPSK] RDA = () is, due to
the assumptions and simplifications necessary to express BPSK1RDA in terms of , only an approximation of the true and exact correlation BPSKl RDA = () The better () approximates true(), the more accurate is the estimated SNR value BPSkXRD -ER in accordance with the invention
In the above-calculated case of BPSK over a real channel, the estimated deviation function  cannot be inverted to a closed form inverse function, so numerical calculation is necessary However, for simplification, the estimated deviation functions can be approximated by a hyperbolic function  which is easy to invert (Formula Removed)
Then the inventive estimated SNR value can be calculated as
(Formula Removed) In Fig 2, (Figure Removed) the function 'as well as the function
(Formula Removed) which approximates the former function are plotted for
comparison The differences are about 0 2 absolutely and about 10% relatively at maximum Their inverse functions, which are available by mirroring the plots at the bisecting line of the first quadrant (dashed line without symbols), are also indicated in Fig 2
In order to quantify the accuracy of the inventive method for estimating SNR values of modulated communication signals, different SNR estimation methods of the state of the art and according to the invention have been tested
A number N, of tests is performed with every method Each test is done with a disjunctive set of N symbols (or bits, samples) The corresponding modulated communication signals rn of each set, with n running from 0 to 7V-1, have a
known true SNR value y In each test, an estimated SNR value m of the tested set is determined by means of the currently tested method, with m test index (or index of tested sets) running from 0 to N, -1 The distribution of the
estimated SNR values m as compared with the true SNR value y is analysed by calculating a normalized mean square error (NMSE) of m
(Formula Removed)
The NMSE values are, for each method, a function of the true SNR value and a function of the number of samples N of each set
Test results are plotted in Fig 3a The abscissa shows the true SNR  in dB,
and the ordinate shows on a logarithmic scale the NMSE values of estimated SNR values for three different methods, i e standard RDA maximum likelihood estimation of the state of the art, inventive RDA-ER maximum likelihood estimation with the estimated deviation function  as in Fig 2, and inventive RDA-ERHA maximum likelihood estimation with the estimated deviation function  as in Fig 2 For the diagram, all in all 106 SNR estimations (tests) have been calculated, with N =100 samples per SNR estimation
For low SNR values (0 dB and less), the NMSE values of the inventive RDA-ER and RDA-ERHA methods are much lower than the NMSE values of the state of the art standard RDA method In other words, the inventive methods are more accurate in this range In particular, at-10 dB and -5 dB, the inventive methods are about 10 times more accurate than standard RDA In said range, RDA-ER NMSE values are about half of the RDA-ERHA NMSE values For higher SNR values (5 dB and above), all three methods are roughly equally accurate
For Fig 3b, the same tests as for Fig 3a have been performed, but with N =1000 samples per SNR estimation The relative difference in accuracy between state of the art RDA on the one hand and inventive RDA-ER and RDA-ERHA is even higher, showing the improvement by the inventive method
For Fig 4a, the same tests as for Fig 3a have been performed, with N =100 samples per SNR estimation again The NMSE values of estimated SNR values

are plotted for the standard RDA maximum likelihood estimation method of the state of the art, the inventive RDA-ER maximum likelihood estimation method, the Iterative method of the state of the art, and the Kurtosis method of the state of the art The inventive RDA-ER method has the lowest NMSE values, indicating the highest accuracy, over a very broad SNR range In the range of 0 dB to 5 dB, the Iterative method is roughly equal to the inventive RDA-ER method
For Fig 4b, the same tests as for Fig 4a have been performed, but with N =1000 samples per SNR estimation For low SNR values (0 dB and less), the inventive RDA-ER method, the Iterative method and the Kurtosis method are equally accurate For SNR values of 10 dB and above, the inventive RDA-ER method clearly outperforms the Iterative method Moreover, the inventive RDA-ER method outperforms the Kurtosis method between 0 dB and 15 dB
In summary, the inventive SNR estimation method has been tested for a BPSK channel over a real AWGN channel It outperforms or is at least equal to known blind SNR estimation algorithms The inventive method can easily be used with other signal modulations over real or complex channels The inventive method requires only limited effort (a little more than the well-known maximum likelihood data assisted estimation), in particular, it does neither need iteration nor decoding/re-encoding of protected data Finally, to avoid the storage and the handling of the optimal interpolation curve   , a hyperbolical approximation is available which allows instantaneous computation with only minor performance degradation

Claims
1. Method for estimating the signal to noise ratio (=SNR) () of a
modulated communication signal (rn) including a data symbol
component (sn) and a noise component {nn),
characterized in that
an intermediate SNR value (RDA) of the modulated communication
signal is derived from a data assisted maximum-likelihood estimation, the assisting data not being known in advance but being reconstructed from samples of the modulated communication signal (r„),
and that an estimated SNR value (RDA-ER) is determined by a controlled non-linear conversion of the intermediate SNR value (RDA).
2. Method according to claim 1, characterized in that the controlled nonlinear conversion is performed by a correction function , with RDA-ER = (RDA) , wherein the correction function  is the inverse function of an estimated deviation function , with the estimated deviation function  approximating a true deviation function  true correlating the deviation of RD from , i.e. RDA =) and
3. Method according to claim 2, characterized in that is chosen such that
() = true () for large numbers of N, i.e. N → ∞, with N being the
number of samples of the modulated communication signal (rn) being processed.

4. Method according to claim 2, characterized in that (Formula Removed)
5. Method according to claim 2, characterized in that  is applied by means of an approximation table.
6. Method according to claim 2, characterized in that (Formula Removed)
7. Method according to claim 1, characterized in that the number N of samples of the modulated communication signal (rn) being processed is equal or less than 500, preferably equal or less than 100.
8. Computer program for estimating the signal to noise ratio() of a modulated communication signal (rn) according to claim 1.
9. Receiver system for estimating the signal to noise ratio () of a modulated communication signal (rn) according to claim 1.
10. Apparatus, in particular a base station or a mobile station, comprising a computer program according to claim 8 and/or a receiver system according to claim 9.