GB2436652A

GB2436652A - Dynamically adaptive multi-stage filter

Info

Publication number: GB2436652A
Application number: GB0606537A
Authority: GB
Inventors: David Lomas; Paul Naish
Original assignee: Matsushita Electric Industrial Co Ltd
Current assignee: Panasonic Holdings Corp
Priority date: 2006-03-31
Filing date: 2006-03-31
Publication date: 2007-10-03
Anticipated expiration: 2026-03-31
Also published as: GB0606537D0; GB2436652B

Abstract

Radio receivers, such as DAB receivers, require filters which are used to discriminate between wanted radio stations and unwanted (interfering) stations. Close channel frequency spacing places high performance requirements on the selectivity of the radio receiver. However, when using on-chip active filters, increased selectivity requires higher power consumption, but is not necessary for all reception conditions. In the present invention, the order of the selectivity filtering is dynamically selected according to reception conditions, with consequent power savings. This is beneficial to battery life in portable applications. The method involves the measurement of signal level (RSSI) before and after the filter. RSSI measurements are taken as filters of different orders are switched into circuit. The set of measurements is compared with a set of templates to ascertain the correct filter to meet the specification for sensitivity in the presence of adjacent channel interference. The comparison algorithm performs a 2-dimensional search over all combinations of interferer distance and power level. A decision can be made on the basis of rms error.

Description

Dynamically Adaptive Multi-Stage Filter The present invention relates

to a dynamically adaptive multi-stage filter whose characteristics can be varied according to prevailing conditions. The invention also relates to a radio receiver having such a filter.

Radio receivers require filters, which are used to discriminate between wanted radio stations and unwanted (interfering) radio stations. Close frequency spacing places high performance requirements on the selectivity of a radio receiver. However, when using on-chip active filters, such as those used in digital audio broadcast (DAB) receivers, increased selectivity requires higher power consumption and is not necessary for all reception conditions. It would therefore be advantageous to be able to use this high selectivity only when it is required.

Under normal circumstances a radio receiver has no information concerning the presence, frequency relationship or relative signal strength of any unwanted adjacent interfering signals and so selectivity filters have to be optimised for worst case conditions only. There are a number of considerable advantages to be gained by the use of an adaptive filter if an adequate method of determining the optimum characteristic can be determined.

US 2004/0142670 Al discloses a dynamically programmable linear radio receiver.

The receiver is coupled to a jammer (interference) detector for detecting the presence ofjamming and the receiver is controlled based on the results of the jammer detector, 50 that when no jammers are present the power consumption of the receiver can be reduced. In this prior art it is suggested that in the digital domain, the coefficients of a programmable digital filter may be adjusted to compensate for lower power consumption.

The present invention provides a method of operating a dynamically adaptive filter in order to retrieve a desired signal from a received signal containing the desired signal with interference, the method comprising: a) applying the received signal to the filter, b) measuring the received signal strength before and after filtering, c) changing the filter order and measuring the received signal strength after filtering, and d) setting the optimum filter order based on analysis of the measurements of signal strength.

The invention is based on the recognition that a receiver does not need to be as selective in the absence of an interferer. With the invention, a receiver becomes self optimised.

Thus, in contrast to the prior art cited above, measurements of signal strength are determined for two or more filter orders in order to determine which filter order is appropriate for the prevailing conditions. Received signal strength indicators (RSSI) before and after the filters are used to estimate the relative power of an adjacent channel interferer and the estimate is used to power down redundant filter stages.

In the specific example to be described in more detail below, the signal strength measurements are compared with stored data, for example in a look up table, in order to determine the appropriate filter order. The stored data consists of measurements of filter response, for two or more filter orders, to test signals having various levels of interference. The test data would typically be generated by simulation. In some cases it would be immediately apparent which filter order is required from the signal strength measurements alone. Therefore the comparison might not he carried out for all signals.

The test signals may include signals having the same interferer at various power levels relative to the wanted signal and/or signals having an interferer at various frequency spacings relative to the wanted signal. With this information in the lookup table, it is possible to decide on the nature of any interference in the received signal in terms of power and/or proximity to the wanted signal.

A typical filter will have three stages all three of which will be used in the determination of the appropriate order. The preferred method of comparing the live received signal strength measurements to the data in the lookup table is equivalent to curve fitting and involves least mean squares calculation.

The invention is particularly suitable for digital radio reception, channel selection and demodulation, and in particular the channel selection filter immediately after conversion to intermediate signal and prior to demodulation.

As well as improving battery life the invention will have the added benefit of improving extreme range reception conditions in a radio receiver by increasing its sensitivity.

A specific example of a method according to the invention will now be described in more detail with reference to the accompanying drawings in which: Figure 1 illustrates the basic topology of a programmable order channel select filter to which the method of the invention may be applied.

Figure 2 shows a typical bi-quadratic element with switched capacitors which may be used to achieve the required response.

Figure 3 shows overall filter responses to an applied signal of given frequency for each configuration or mode.

Figure 4(a), (b) and (c) are three sets of curves as examples of signal strength measurements that may be stored in the lookup table.

Figure 5 schematically illustrates a typical radio system in its entirety showing the relationship of the filter adaptation logic to the rest of the system.

Figure 6 is a flow chart showing the principle process steps of the present invention.

Figure 7 shows a part of the process of the previous figure in more detail.

Figure 8 shows the LMS algorithm in more detail.

Figure 1 shows an adaptive filter topology with three filter stages comprising a first order filter 1, also called an integrator, with bi-quadratic second and third stages, 2 and 3. Analogue switching, generally indicated by numeral 4, is provided in order to selectively bypass the higher order stages 2 and 3, if necessary, or even the entire filter if there is no interference at all. Within each of the filter stages there are switched frequency selection components that adapt the respective stage to supply the correct response.

Figure 2 shows that the same results can be achieved using a bi-quadratic element with switched capacitors. Other topologies are possible.

The theory behind the present invention will now be explained: We write the filter transfer characteristic in the Laplace notation as a function of 5: H(s)=f(s) We illustrate the idea by determining the optimum order for a filter of the form H(s)= k0 P(a) Where P(w) is a simple polynomial in (D. In general the numerator may also be a polynomial in o.

Highly selective filters have high order polynomials, which may be factorised into the product of a set of low order filters H(s) = Ho(s).H1(s).H2(s) Each of these stages will require power, and hence increase battery drain in the following manner P(tot) = P(0) + P(1) + P(2).. So

P(tot) E P(n) over all stages Similarly, assuming unity gain at each stage, the noise contributions also sum linearly N(s) = EN(n) over all stages So we can see immediately there is a conflict between increasing the number of stages to improve selectivity but decreasing the number of stages to improve power supply demand and to improve the noise performance of the receiver.

Internationally agreed standards dictate that the receiver must meet minimum sensitivity requirements in the presence of given levels of adjacent channel interferers but in the absence of interferers it is perfectly acceptable and feasible to relax these filtering requirements by bypassing some stages and adjusting the polynomial coefficients of the remaining stages. This decreases the power consumption and improves receiver sensitivity.

Note that the time constants of the remaining active filter elements have to be modified to retain an optimum filter of lower order and the additional circuitry required to implement this is shown in Figure 2. The new filter pass-band characteristics are shown in Figure 3, as overall filter responses to an applied signal of given frequency for each of the configurations or modes. This clearly shows the change in response for an adjacent filter of different order.

However there is no a priori method or coding within the signal or within the radio itself to determine the optimum filter order and thus all methods and techniques must be post priori, i.e. after reception. This is where the importance and uniqueness of this application lies.

We can illustrate the functionality of the auto-optimisation algorithm by considering a filter switched between two arbitrary orders; m &, n Since the filters have different orders, they will have different responses at the frequency of the first or subsequent adjacent channels. The further away in frequency that an interferer is, the greater the change in response is.

By application of the filter to the applied signal and switching through all the available modes (see Figure 1 for the switch topology) we obtain a set of measurements of the RSSI, both pre & post filter and, from comparison with a set of templates, imply the correct filter order to meet the specification. In the following, by means of three examples we illustrate specific cases from which a decision can be inferred concerning the nature of the interferer.

We illustrate in Figure 4a as the first example, the case of a strong interferer as this is the most intuitively accessible i.e. adjacent signal strength > wanted channel signal strength. The different curves show the effect of interference on the adjacent channel, the channel next to the adjacent channel and so on. The spacings are based on the channel separation for DAB signals. For everywhere except Canada this is 1736kHz, and for Canada the spacing is 1744kHz. The difference is small such that the same templates should be universally applicable.

The curves are normalised so that the RSSI at the input of the filter is at a constant level. This is the function of the automatic gain control. (As a footnote here, the primary function of the automatic gain control is to hold the signal level constant at the input of the filter to optimise its dynamic range; the normalisation of the RSSI for this algorithm is a secondary function) Plotting the RSSI against filter order we observe that this forms a family of curves, the closer to the wanted signal the interferer is, the less increasing the filter's order changes the RSSI measured at the output of the filter.

The graphs of figures 4a and 4b demonstrate the same process with successively lower interferer received signal strengths.

From inspection of the curves, it can be seen that the problem of identifying the relative value and position of an interferer then reduces to that of matching the measured curve against a set of known curves. There are a number of known methods in other fields that can be used to perform this function. The best known is the least mean squares' or LMS method, from which the root mean square error between the measured and stored curve is determined for each of the curves held in store. The minimum error then indicates the closest curve and therefore the best estimate of adjacent frequency and power.

An alternative technique may be to use state machine based decision methods such as the Hidden Markov Model.

Descrintion of algorithm At present, a working spreadsheet prototype has been demonstrated for a single interferer of unknown amplitude and frequency. An input vector is formed from a set of numbers representing the output from the RSSI corresponding to all the available filter orders.

X = {x:x is an RSSJ value obtained flerJIltering} With x0 = no filter, (i.e. the null filter; by-pass mode) x1 = power level measured by application of a first order filter x2 = as above but with a second order filter Etc.

Giving X = {x0, x1, x2, x3 etc} Although the mathematics describes a complete set of filter orders, the topological description (Figs. 1 and 2) describes our implementation, which has filter orders {0,1,3,5}. Using such a subset does not require any modification of the algorithm.

To simplify the mathematics & improve numerical accuracy, we normalise the measured set of values so that the first value, xO, is unity, thus obtaining a mapped set of measurements, X', where X' = {x', x'1, x'2, x'3 etc}, where x'0 1, x'1, = x1/ x0 etc There also must exist a similar set of stored reference vectors, Y(m,n) against which the input, RSSI vector can be compared. This set could be pre-calculated, i.e. stored as a look up table, or calculated on-the-fly' from the transfer functions of the filter orders. The two dimensions, m & n, represent the adjacent interferer's channel position and the interferer's amplitude respectively. So

y0(m,n)= Pre-calculated response of system with no filter (by-pass mode) to an adjacent interferer m channels away of amplitude n y1(m,n)= Pre-calculated response of system with a first order filter to an adjacent interferer m channels away of amplitude n y2(m,n)= Pre- calculated response of system with a second order filter to an adjacent interferer m channels away of amplitude n y3(m,n)= third order as above Etc.

These are also normalised so that yo(m,n)=l The Euclidean distance is defined as: for coordinates of points dim = {(x1 _xn.)2J x1 and xm on the axes of a k-dimensional space.

(reference 1) Using the above expression, the Euclidean distance between X & any stored vector Y(m,n) now becomes E(m,n)=I((xo_yo(m,n))2 + (xi-y1(m,n))2 + (x2-y2(m,n))2 + (x3-y3(m,n))2) Although for the purpose of searching for the best match, taking the square root is not necessary, simplifying the formula to E(m,n)2= (xo-y0(m,n))2 + (x1-y1(m,n))2 + (x2-y2(m,n))2 + (x3-y3(m,n))2 Since the data values are normalised with the first terms of both arrays equal to unity, the expression simplifies to (E'(m,n))2 = + (x1 -y1' (m,n))2 + (x2'-y2' (m,n))2 + (x3'-y3' (m,n))2 Normalisation also achieves some increase in numerical accuracy as only differences are processed.

The essence of the algorithm is therefore a 2-dimensional search over all combinations of interferer distance and interferer power levels, searching for a minimum value. A decision regarding the position & strength of a possible interferer can then be made on the basis of either achieving an RMS error (Euclidean Distance) less than a preset threshold (which then defines the acceptable error and hence overall accuracy of the algorithm) or by an exhaustive search over all possible combinations of interferer amplitude and position. The exhaustive search is the more thorough and rigorous method but has the disadvantage of higher computational overhead.

As can be seen from the description, the program searches over all combinations of adjacent interferer and amplitude by means of two nested loops. The least RMS error (Euclidean distance) and corresponding amplitude & position is retained and updated if a better combination of interferer position & amplitude is found.

Searches for multiple interferers may be taken by an extension of the method into a multidimensional search space by the deeper nesting of the nested loops.

The preferred filter for use in the invention will have the following characteristics: * Dynamically re-configurable filter * Continuous Gm-C filter-Chebyshev Low Pass Filter * High linearity, High IP1dB * Characteristic digitally tuned against crystal reference (tolerance within +12%) * Order dynamically switched between 5th, 3rd, 1st and Bypass mode.

* Order of filter dynamically adjusted to account for level of adjacent channels * Filter Accuracy, -this is ensured by filter tuning to within +1-2% * Ability to measure the power before and after the filter -this is achieved using wide band RSSI level indicators as shown * Ability to switch the filter characteristic without excessively distorting the wanted signal, achieved by switching the tuning components and by powering up the next filter stage prior to an order change The powering down of filter stages not required can save up to an additional 20mW of power.

Optional features: 1) The process of auto-optimisation during the redundant period of a transmission frame when data not required by the receiver is being transmitted.

2) The process of pre-powering up of Gm stages before use in order that the minimum of disturbance be applied to the required received data prior to demodulation.

Figure 5 shows a typical radio system in its entirety showing the relationship of filter adaptation logic to the rest of the system. Figure 5 shows an example of the context of the invention. Drawn is a typical radio system in its entirety showing the described filter adaptation logic 20, the channel select filters 25 (both I & Q), RSSI measurement at the input and output of filters 25, and the auto-calibration for filter tuning. Auxiliary components that are not essential to the invention, but are still required to implement a full receiver such as mixers, gain control elements etc are also shown. Note the signal flow, with the RSSI continually measuring the power levels, the filter adaptation logic 20 interpreting the values and then switching filter order appropriately, via filter tuning logic 30.

The filter should be allowed to power up & down quickly enough to allow calibration but not to disturb the received signal.

A key aspect of the invention is the operation of the Filter Adaptation Control Logic whose function is to take two or more measurements of both input & output RSSI with different filter orders, and implement the described mathematical relationships.

Filter order modification is achieved by successively bypassing the potentially redundant filter stages and to examine the values of the Received Signal Strength Indicators. From absolute values of measured signal strength, implications can be made about the strength & position of adjacent channels and decisions can be made about the optimum filter order.

To summarise we can therefore see that the process of optimising can be seen as 1) Accurately measuring the input RSSI 2) Accurately measuring the change in output RSSI upon a change in filter order 3) Using the data to set an appropriate filter order for that reception condition.

The filter order control processes will now be described in more detail with reference to Figures 6, 7 and 8. Figure 6 shows the principle process steps. After the start the first step is to obtain the RSSI value for each of the filter orders. An algorithm such as LMS is used to estimate the amplitude and frequency position of an interferer and on this basis an appropriate filter order is selected. A linear process is illustrated with a single start and stop. The actual process would be continuous with a delayed link from finish to start. The delay time would be optimised for the system as part of the system design.

Figure 7 shows the first step of Figure 6 in more detail. In the process of Figure 7, firstly the automatic gain control is set so that the RSSI at the input of the filter measures 1. The filter order is set to 0. The RSSI is then measured at the filter output and the RSSI is measured with each filter order. The results are supplied to LMS algorithm.

Figure 8 shows the LMS algorithm in more detail. As the Figure shows, this exemplary process is an iterative process to find the combination of the amplitude and frequency that produce the lowest RMS error. The determination of amplitude and frequency is then used to select an appropriate filter order.

As part of this application and general ongoing development, we include a class intelligent algorithms' whose operation is not based on a full search of the Euclidean distance to decide interferer position and power level but also include a method of quickly deciding on an appropriate filter order. For example, a small (normalised) value of x3 or x5 is an immediate indicator of a large interferer at least two or more channels away from the wanted carrier. This immediately indicates a high order filter is required without further analysis. With this search space eliminated, the algorithm then is only required to search for strong carriers near to the wanted carrier.

Claims

CLAIMS: 1. A method of operating a dynamically adaptive filter in order

to retrieve a desired signal from a received signal containing the desired signal with interference, the method comprising: a) applying the received signal to the filter, b) measuring the received signal strength before and after filtering, c) changing the filter order and measuring the received signal strength after filtering, and d) setting the optimum filter order based on analysis of the measurements of signal strength.

2. A method as claimed in claim I comprising: providing a data store containing measured values of output signal strength for each filter order in response to test signals representing the wanted signal with various levels of interference, and comparing measurements of received signal strength with data in the data store to select an appropriate filter order.

3. A method as claimed in claim 2 in which the test signals used for the data store include signals having the same interferer at various power levels relative to the wanted signal.

4. A method as claimed in claim 2 or 3 in which the test signals used for the data store include signals having an interferer at various frequency spacings relative to the wanted signal.

5. A method as claimed in claim 2, 3 or 4 in which the comparison step includes determining the nature of an interferer in the received signal in terms of power and/or proximity.

6. A method as claimed in any of claims 2 to 5 in which the comparison at step (d) involves a least mean squares calculation.

7. A method as claimed in any of claims 2 to 6 in which step (d) involves a two dimensional search over combinations of interferer distance and power level to determine a minimum value and hence estimate the distance and power level of interference in the received signal.

8. A method as claimed in claim 7 in which a possible interferer is identified when the result of the comparison is below a threshold value.

9. A method as claimed in claim 7 or 8 extended to a multi-dimensional search in order to identif' multiple interferers.

10. A method as claimed in any preceding claim which step (c) is carried out for three different filter orders.

11. A method as claimed in any preceding claim in which the filter stages are powered up before taking signal measurements.

12. Operation of the channel selection filter of a DAB receiver according to the method of any preceding claim.

13. Operation according to claim 12 in which the filter order is set during the redundant period of a transmission frame.

14. A dynamically adaptive filter configured to operate according to the method of any preceding claim.

15. A digital audio broadcast receiver having as channel select filter a filter as claimed in claim 14. ft

Amendments to the claims have been filed as follows I. A method of operating a dynamically adaptive filter in order to retrieve a desired signal from a received signal containing the desired signal with interference, the method comprising: a) applying the received signal to the filter, b) measuring the received signal strength before and after filtering, c) changing the filter order and measuring the received signal strength after filtering, and d) using the received signal strength measurements to estimate the power and/or proximity of an interferer e) setting the optimum filter order based on analysis of the measurements of signal strength I) powering down any filter stages not required.

2. A method as claimed in claim 1 comprising: providing a data store containing measured values of output signal strength for each filter order in response to test signals representing the wanted signal with various levels of interference, and comparing measurements of received signal strength with data in the data store to select an appropriate filter order.

3. A method as claimed in claim 2 in which the test signals used for the data store include signals having the same interferer at Various power levels relative to the wanted signal.

4. A method as claimed in claim 2 or 3 in which the test signals used for the data store include signals having an interferer at various frequency spacings relative to the wanted signal.

5. A method as claimed in claim 2, 3 or 4 in which the comparison step includes determining the nature of an interferer in the received signal in terms of power and/or proximity.

6. A method as claimed in any of claims 2 to 5 in which the comparison at step (e) involves a least mean squares calculation.

7. A method as claimed in any of claims 2 to 6 in which step (e) involves a two dimensional search over combinations of interferer distance and power level to determine a minimum value and hence estimate the distance and power level of interference in the received signal.

8. A method as claimed in claim 7 in which a possible interferer is identified when the result of the comparison is below a threshold value.

9. A method as claimed in claim 7 or 8 extended to a multi-dimensional search in order to identify multiple interferers. S..

10. A method as claimed in any preceding claim which step (c) is carried out for three different filter orders.

11. A method as claimed in any preceding claim in which the filter stages are powered up before taking signal measurements.

12. Operation of the channel selection filter of a DAB receiver according to the method of any preceding claim.

13. Operation according to claim 12 in which the filter order is set during the redundant period of a transmission frame.

14. A dynamically adaptive filter configured to Operate according to the method of any preceding claim.

15. A filter as claimed in claim 14 having two or more stages with switching for selectively by passing the higher order stage(s).

16. A digital audio broadcast receiver having as channel select filter a filter as claimed in claim 14 or 15. s. S *