Methods and arrangements for adaptive thresholds in codec selection

(54) Title: Procedures and arrangements for adaptive thresholds in codec selection (56) Published publications: 'Rate adaptive speech coding for universal multimedia access' by Homayounfar, K.2003 (47) Abstract:

A method and arrangement for dynamic adaptation of thresholds used to select a codec mode to be used is presented. Thresholds are adjusted (240) in response to the current received signal quality. An estimate of the actual signal quality received prevails (230) on which the adaptation is based. The present invention can be applied either to the mobile terminal side or the network side, to act on the uplink and / or downlink. The thresholds can be modified on the receiver side, or when operating in the network and working on the downlink, the threshold adjustment can be initiated in the terminal.

<img file="SE528213C3_D0001.tif" />

FINAL

Fie. 5

528

SUMMARY

A method and arrangement for dynamic adaptation of thresholds used to select a codec mode to be used is presented. Thresholds are adjusted (240) in response to the current received signal quality. An estimate of the actual signal quality received prevails (230) on which the adaptation is based. The present invention can be applied either to the mobile terminal side or the network side, to act on the uplink and / or downlink. The thresholds can be modified on the receiver side, or when operating in the network and working on the downlink, the threshold adjustment can be initiated in the terminal.

(Fig. 5)

528

4 · 4 V

TECHNICAL FIELD

The present invention relates generally to communication systems using multi-codecs and methods therefor, and in particular to methods and devices using selection thresholds.

BACKGROUND

Voice codecs that use Adaptive Multi Rate (AMR) [1,2] enable the provision of excellent speech quality and at the same time provide a way forward towards cellular networks with the best known technology, spectral efficiency and high capacity. A straightforward way to characterize the benefits of AMR speech codecs is that the robustness to interference and noise in radio networks is increased and that this advantage over other non-adaptive speech codecs can be transposed in many different ways, e.g. by increasing speech quality or improving spectral efficiency. AMR Code Organizations are standardized by 3GPP for both the Global System for Mobile Communications (GSM), which is currently the world's most widely used cellular technology, as well as for Broadband Code-Shared Multiple Access (WCDMA, Wideband Code Division Multiple Access). .

Narrow banding AMR consists of eight codec modes with different source bit rates, from 12.2 kbps down to 4.75 kbps. For AMR in GSM, a number of codec modes are assembled into a predefined Active Codec Set (ACS), which is fixed during a call. The channel coding level is adjusted while the overall bit rate is constant. Consequently, the lower the source bit rate, the more robust the codec against bit errors. For codec mode customization, the receiver side performs link quality measurements on the incoming channel which gives a quality indicator (QI), which is

528 213 is defined as an equivalent carrier-to-interferer ratio (C / I) (3]. The QI is then compared against a set of fixed, predefined thresholds to determine which codec mode to use.

SUMMARY

In order to obtain the best possible speech quality in AMR codecs, it is important to choose the codec modification thresholds correctly. However, obtaining a QI that accurately reproduces the voice quality for all radio conditions, frequency hopping schedules and network configurations can be quite complicated. Furthermore, conditions can vary over time. There may also be performance variations between different receiver units, both in terms of actual performance and QI estimation. This means that it is likely that even well-chosen adaptation thresholds will not be optimal at all times. Fixed thresholds can be sub-optimal for the present conditions by either being too high or too low. In the case where the thresholds are too high, a switch from a less robust codec mode to a more robust codec mode will be initiated earlier than the radio conditions require. This will cause a slight degradation of speech quality due to the lower inherent speech quality of the more robust code mode. A more serious problem arises when the thresholds are too low, which causes the switch from the less robust mode to occur too late. This can significantly increase the errors on the radio link and in turn cause a serious degradation of speech quality. Thus, since both cases lead to speech quality reductions, the two should be avoided.

A general object of the present invention is to provide improved methods and devices for selecting codec modes in multi-codec systems. It is a further object of the present invention to provide methods and devices which provide selection of codec modes that are less sensitive to differences in radio conditions or estimation thereof.

-v η

213

The above-mentioned objects are achieved by methods and devices according to the appended claims. In general, the proposed solution is to use thresholds that are adaptable in response to the received signal quality present. A preferred algorithm for trigger matching is quite general and can be applied either on the terminal side or on the network side, acting on the uplink and / or downlink. The thresholds can be modified on the receiver side, or, when operating in the network and acting on the downlink, the threshold adjustment can be applied in the terminal. The present invention is intended to work with AMR and AMR-like speech and audio codecs, e.g. AMR-WB and AMR-WB +, regardless of the special radio access method, but can also work with other code-selection techniques based on link quality thresholds.

An advantage of the present invention is that thresholds for selecting suitable codec modes are kept adapted to prevailing radio conditions and estimated characteristics, which increases the overall average speech quality.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention, together with further objects and advantages thereof, can best be understood by references to the following description made together with the accompanying drawings, in which:

FIG. 1 is a schematic block diagram of a wireless communication system;

FIG. 2A is a schematic block diagram of a typical downlink transmission AMR codec system;

FIG. 2B is a schematic block diagram of a typical AMR codec system for uplink transmission;

FIG. 3 is a diagram illustrating relationships between link qualities and speech qualities for different codec situations;

FIG. 4 is a flowchart of main steps for an embodiment of a method according to the present invention;

523 213

FIG. 5 is a flowchart of main steps for another embodiment of a method according to the present invention;

FIG. 6 is a flowchart of main steps for a further embodiment of a method according to the present invention;

FIG. 7 is a diagram illustrating simulation results according to embodiments of the present invention;

FIG. 8A is a diagram illustrating a threshold adjustment principle used in one embodiment of the present invention;

FIG. 8B is a diagram illustrating a threshold adjustment principle used in another embodiment of the present invention;

FIG. 8C is a diagram illustrating a threshold adjustment principle used in yet another embodiment of the present invention;

FIG. 8D is a diagram illustrating a threshold adjustment principle used in a further embodiment of the present invention;

FIG. 8E is a diagram illustrating a threshold adjustment principle used in yet another embodiment of the present invention;

FIG. 9A is a schematic block diagram of main parts of an embodiment of a downlink transmission codec system according to the present invention;

FIG. 9B is a schematic block diagram of main parts of another embodiment of a downlink transmission codec system of the present invention;

FIG. 9C is a schematic block diagram of main parts of an embodiment of an uplink transmission codec system according to the present invention; and

FIG. 9D is a schematic block diagram of main parts of another embodiment of an uplink transmission codec system according to the present invention.

528 213

DETAILED DESCRIPTION

Fig. 1 schematically illustrates a wireless communication system 1. A mobile terminal 20 communicates with a base station 10. Signals transmitted from base station 10 to mobile terminal 20 are referred to as downlink signals 12 and signals transmitted from mobile terminal 20 to base station 10 are referred to as uplink signals 22. For example, radio conditions are determined. . of the distance between the transmitter and the receiver, and the topology of nature in the surroundings of the signal path. Radio conditions are also affected by interfering radio signals. Another base station 19 may e.g. transmit signals 11, 13 which may interfere with the uplink 22 and downlink 19 signals. Similarly, another mobile terminal 29 may also provide interfering signals 21, 23.

In an AMR approach, different link conditions require being able to use different codecs to ensure a certain speech quality. Link conditions are quantized as a link quality LQ, typically related to a measure of C / I. Many different methods of estimating C / I are available in the prior art, performed at the transmitting and / or receiving end. C / Dimensions can thus either be measured directly or indirectly or reported from elsewhere in the system. Based on LQ, an appropriate codec mode is selected. Typically, such a choice is implemented through a set of thresholds. For LQ values ​​below a first threshold, a first codec mode is applied, typically the most robust available. Between the first threshold and a second higher threshold, a second codec mode is applied, and so on. By introducing n thresholds, one can choose between n + 1 codec modes.

Fig. 2A illustrates a typical AMR codec system for downlink transmission. A code select unit 39 in a base station 10 comprises an AMR codec 30. The AMR codec 30 includes in the present embodiment four codecs 31 AD. A modem switch 32 connects an input signal terminal to one of the codec modes 31 AD. The modem switch 32 is controlled by a threshold comparator 35 in a switch controller 33. The switch controller

528 213 also includes a link quality provider 34, which provides an estimate of LQ for the link intended to be used. Typically, the link quality provider 34 receives an LQ estimate from the mobile station 20. The threshold comparator 35 determines between which thresholds LQ is located and selects a codec mode accordingly. The parts in the figure are typically functional parts and can very well be implemented with software in one and the same processor.

The signal to be transmitted is coded using the selected code mode and transmitted over the downlink 12.

The base station 10 also informs the mobile station 20 about which codec mode is used. This can be done either in a separate message using any type of control signaling or in conjunction with the encoded signal itself. The mobile station 20 receives the information about which codec mode is used in a switching controller 45. The mobile station 20 includes an AMR decoder 40, which in the present embodiment comprises in turn four decoder modes 41A-D and a modem switch 42. In response to the information on which codec mode used in the transmitter, modem switch 42 selects the appropriate decoder mode 41 AD. The decoded signal is then forwarded for further processing, here represented by a unit 49.

Fig. 2B illustrates a typical AMR codec system for uplink transmission. A code select unit in a mobile terminal 20 includes an AMR codec means 30. In the present embodiment, the AMR codec 30 includes four codec modes 31 AD. A modem switch 32 connects an input signal terminal to one of the codec modes 31 AD. The modem switch 32 is controlled by a switch controller 33. Typically, the mobile station 20 receives in the switch controller 33 the information about which codec mode to use from the base station 10. In alternative embodiments, the switch controller 33 can make such decisions itself. The parts in the figure are typically ο 4 ς

4.

functional parts and can very well be implemented with software in the same processor or in several processors.

The signal to be transmitted is coded using the selected code mode and transmitted over the uplink 22.

The base station 10 comprises an AMR decoder 40, which in the present embodiment comprises four decoder modes 41A-D and a modem switch 42. In response to the information on which codec mode to be used in the mobile station 20, a switch controller 45 controls the modem switch 42 to select the appropriate decoder mode 41 AD. The decoded signal is then forwarded for further processing.

In addition, according to the present invention, devices, systems and methods according to the principles presented above are provided with means for adjusting thresholds based on prevailing received speech quality.

Fig. 3 is a diagram illustrating in the left part thresholds T1-T5 for codec mode selection. In the current GSM-AMR, a maximum of four codecs can be used in each ACS, which means that three thresholds are used. However, the present invention is not limited to GSM-AMR and works with any number of modes. Fig. 3 therefore represents a generalized configuration having 6 modes and 5 thresholds. For LQ values ​​between these thresholds, different codec modes M1-6 should be selected. The thresholds T1-5 are determined to, at least theoretically, provide numbers within a certain target quality range Q, as illustrated by the dashed region in the right part of the diagram. If the thresholds Tl-5 are determined appropriately and if the link quality value on which the choice is based is correct, then an actual link quality, e.g. represented by the dashed line 100 to provide a speech quality 101 within the range Q.

Assuming that the link quality is overestimated, ie that a provided estimate 102 for a link quality is higher than that

213 the actual 100, the choice of codec mode should not be optimal. In Fig. 3, the measured link quality falls above the T2 threshold, which means that codec mode M3 is chosen instead of the optimal choice M2. As a consequence, the average quality of the transmitted number will be raised over the interval Q, represented by the dashed line 103.

In such a situation, the threshold T2 would preferably be adapted as a kind of calibration to the, somewhat incorrect, link quality estimate in the receiver. By increasing the T2 threshold to be located at the level represented by dotted line 104, a proper selection of codec mode M2 ​​will be performed, which will result in a speech quality within the range Q.

It can be noted here that if the inaccuracy in link quality estimation is systematic, the same error will exist at all thresholds, and preferably all thresholds could be adjusted by the same amount.

Likewise, if the assumed relationship between a certain combination of link quality and codec mode, and the resulting speech quality is somewhat incorrect, e.g. if a theoretical oversimplification is done, an adjustment of a threshold can blur the result to the desired speech quality range. In cases where such relationship errors are systematic, a simultaneous adjustment of all thresholds might be preferred. In other cases, individual adjustments to individual thresholds are more appropriate.

An embodiment of a method according to the present invention is illustrated in Fig. 4 on a general level. The procedure begins in step 200. In step 220, link quality measurements are provided. In step 222, a codec mode is selected to be used among a set of available codec modes by comparing the measured link quality with a set of thresholds. An estimate of the current received signal quality is provided in step 230. In step 237, it is determined if the signal quality estimation is outside a predetermined signal quality interval for the code mode used. If the signal quality estimation is within the range, no threshold adjustment is made, or alternatively a zero-size threshold adjustment is made, and the procedure proceeds to step 299. If the signal quality estimate is out of range, the procedure proceeds to step 240, in which the set of thresholds is dynamically adjusted in response to the estimate of the current received signal quality. The procedure ends in step 299. Although the procedure is illustrated as a single row of steps, the actual procedure is typically repeated a number of times, as illustrated by the dashed arrow 250.

Also note that the threshold adjustment steps can be separated from the actual selection steps, and the execution of the different steps can be repeated independently of each other.

A particular embodiment of an threshold adjustment step algorithm is presented in connection with Fig. 5. The adaptive threshold algorithm estimates the speech quality of the received link and compares the estimate against given speech quality limits for each codec mode. The speech quality could be estimated from frame erasure measures, e.g. FER, bit error rate measures, e.g. RxQual, or objective speech quality measures, e.g. SQI [4] or PESQ [5]. If the estimated speech quality of a given mode is out of bounds, either too good or too bad, it is likely that the associated threshold for switching to the appropriate adjacent codec modes is sub-optimal for the current radio conditions. The algorithm will then modify all codec switching thresholds. One reason for modifying all thresholds instead of only the associated threshold is that it makes it much easier to always maintain the thresholds in a consistent order, ie. that they do not overlap [3]. Such problems are discussed in more detail below.

In a typical case, a single value for speech quality is estimated for the receiving link too noisy for direct use in a threshold adjustment decision. Instead, in the present embodiment, a long-term average value is calculated. The calculation of the threshold adjustment is based on the long-term speech quality and on the achievable speech quality with the current ACS. Due to the average formation process used to obtain longCon η * «*

WAV 4 ». * ·% Ί;

In the time frame quality, several threshold adjustments will usually need to be made.

For an application of a network-based algorithm that works on the downlink signal, the new set of thresholds will be transmitted to a mobile station by stealing a speech frame, which effectively causes a deleted frame. As a consequence, the algorithm should preferably not be allowed to update the thresholds too often. However, simulations show that this is typically not a problem in realistic scenarios.

For an illustration, an algorithm for adaptive thresholds for AMR in GSM has been developed and evaluated. The proposed threshold adjustment is applied to a configuration that works on a webpage and downlink, since it is considered particularly useful. This enables the alignment of AMR performance for mobile stations from different vendors without having to change the mobile station, which would require standardization. Input to the algorithm in this particular embodiment, FER is reported by the mobile station, ie. FER is the speech quality measure as referred to above. The FER value is obtained from the Enhanced Measurement Report (EMR) [6]. As a result, it is a necessary requirement for the algorithm in the present embodiment to work for the mobile station to support the EMR. The EMR includes the number of correctly received frames during the measurement period of 480 ms, ie. 24 speech frames. The FER value is not reported directly in the EMR. What is reported is instead the number of blocks correctly received. However, the FER value can be calculated from this because the base station system (BSS) knows the number of blocks transmitted. The algorithm is intended to be run as soon as an EMR has been received.

The procedure begins in step 200. In step 231, the input data necessary for the algorithm is provided. The algorithm of the present embodiment needs two inputs: the mobile station's FER value and the latest code mode used by the mobile station during the last EMR measurement period. Both inputs can be obtained, directly or indirectly, through the EMR sent from the mobile station.

528 213

If a threshold change has been made, it will take a while before the change has been applied by the MS. This means that one or a few FER measurements, measured immediately after the decision to change the thresholds have been made, will be representative of the old threshold and should therefore not be used for further evaluation. The present algorithm embodiment therefore has a counter, which counts how many EMR reports have been received since the last threshold change. A certain predetermined latency period is defined and the EMRs for which the latency period has not expired are discarded. This is illustrated in Fig. 5 as step 232. In that case, the present algorithm returns immediately without modifying any thresholds.

The number of EMRs to throw away in the present embodiment is specified by a parameter, and its normal value is 1, ie. only the first EMR is discarded immediately after a threshold change. The reason for this is that a threshold change normally takes less than 480 ms to execute.

The FER value from the EMR is stored in step 234 in an array for the subsequent calculation of long-term average value FER. The behavior of the example algorithm can to a large extent be modified by changing how many EMRs to receive before proceeding with the calculation of the threshold adjustment. The length of the FER array is controlled in the present embodiment by a parameter. One possibility is to specify that the entire FER array should be filled before a threshold is changed. Another possibility is to always allow threshold changes regardless of how many EMRs have been received. The first case will henceforth be called ATI and the second case AT2. In extreme cases it can be said that ATI gives big threshold changes but rarely, while AT2 gives small changes but more often. As will be discussed further below, AT2 will also provide threshold changes that basically have fixed values.

Since the threshold changes will be sent to the MS by stealing speech frames, it is clear that making threshold changes too often will impair speech. On the other hand, the normal length of the FER array is so large (its

<img file="SE528213C3_D0002.tif" />

normal length is 60 in the present embodiment) so waiting until it is completely filled before the thresholds change means that the thresholds will remain unmodified for a substantial period of time. It usually takes almost 30 seconds. If the FER estimate in the MS is seriously wrong, this is too long to wait before the thresholds change. Results from simulations show that the thresholds should be allowed to change as often as possible. Nevertheless, the benefits of changing the thresholds as soon as possible are so great that the present algorithm embodiment should be able to do so, but with a small delay period, which prevents the thresholds from changing too often.

In the present embodiment, the number of EMRs that must be received before a threshold change can be made by a parameter whose normal value is 10. In step 235, it is checked if the number of received EMRs is less than this parameter. If not enough EMRs have been received, the procedure returns without modifying the thresholds.

The parameter that controls the number of EMRs needed to perform long-term averaging may also depend on various other parameters. Non-exclusive examples of such parameters are current codec mode, receiver equipment characteristics, time of day and day of the week.

Long-term FER, FERit, is calculated in the present embodiment in step 236 by averaging the entire FER array. Note that the entire FER array is used in this calculation, although not all array elements need be filled. This means that long-term FER will be drastically underestimated when only a few elements in the array have been filled. This is done quite deliberately to avoid having threshold changes that occur too often, and works in parallel with the control described above.

Long-term FER, FERit, is then compared to an upper and a lower FER limit in step 237, and if FERit is below the lower FER limit or above the upper FER limit, the procedure returns without modifying codec mode switch13 thresholds. In the present embodiment, the normal value for the above FER limit is 0.01, ie. 1%, for full rate (FR, full rate) and 0.03, ie 3%, for half rate (HR, half rate).

From FERit, in step 240, a threshold adjustment Θ is finally calculated for all thresholds. In the present embodiment, it is calculated as:

A® = Cj₀¹⁰log

FER "

FER, = C^{FER t}

FER, = C₂²log- ^ '

FER,

There, the FERt is the target against which the present algorithm embodiment aims. Its normal value is 0.0015, ie 0.15%, for FR and 0.003, ie. 0.3%, for HR. The constant C controls the aggressiveness of the present algorithm embodiment, i.e. how quickly it tries to reach the FER target. The value of C will depend on the basis of the logarithm of the equation above. Table 1 gives values ​​for logarithms in base 10, e and 2.

<td></td><td>FR</td><td>HR</td>

<td>CIO</td><td> 3.208</td><td> 4.960</td>

<td>Ce</td><td> 1.393</td><td> 2.154</td>

<td>c₂</td><td> 0.9656</td><td> 1.493</td>

Table 1. Values ​​of the constant C for use with logarithms with different bases.

A corresponding equation for a more general signal quality measure can be formulated as:

* ® = f (SQ_e, SQ,),

528 z: z where ΔΘ is the threshold adjustment, / () is a predetermined function that has two variables, SQ_e is the estimated prevailing received signal quality and SQ, is a measure of received signal quality.

In the present particular embodiment, the threshold adjustment ΔΘ is then rounded off to the nearest higher half decibel value and added to all the present codec switching thresholds.

Note that for AT2, in which we do not wait for the FER array to be filled before calculating the threshold adjustments, but instead try to make a threshold adjustment as soon as an EMR has been received, the threshold adjustment will be almost constant. The reason for this is that a threshold adjustment is calculated as soon as the long-term FER exceeds the FER limit, which means that the FERit will be very close to the FER limit at the time of the threshold adjustment, except for pathological cases. For FR and using the normal value of the FER limit as the value of FERit in the equation further above, we obtain a threshold adjustment ΔΘ - 2.6 dB, or rounded to the nearest higher half decibel value, ΔΘ = 3.0 dB. For HR, we get a rounded threshold adjustment ΔΘ = 4.0 dB.

The upper FER limit was derived from the FER levels at the codec modifier thresholds for all ACSs. The highest FER level for the higher code mode at the downward switching threshold was taken as a reference value. If any time spent in the lowest code mode is neglected, this value is basically the highest FER level that will ever be obtained under optimum settings. If a higher FER level is obtained, the link fitting will select a lower codec mode. This FER reference value was then multiplied by 1.5 and rounded to obtain the upper FER limit.

The lower FER limit is obtained in an analogous way. Here, the highest FER level for the lowest code mode is taken at the upward switching threshold

Z. Μ

Z.

as a reference value. If any time spent in the highest code mode is neglected, this is an upper limit for the lowest FER level ever obtained under optimal settings. If a lower FER level is obtained, the link fitting will select a higher codec mode. This FER reference value was then divided by 1.5 and rounded to obtain it below the FER limit.

Target FER is obtained in a similar way to the lower FER limit. The target FER is then taken approximately as 2 times the reference value for FR, and 4 times the reference value for HR.

The constant C is obtained by doing a linear approximation of the FER-to-C / I ratio in the log domain. The value of C is then taken as the absolute value of the inverted value of the slope of the linear approximation. The slope depends somewhat on the codec mode, so the average over all codec modes is taken (separately for FR and HR). There is a greater dependence on the number of frequencies used for frequency jumps. The more frequencies, the greater the slope. Jump over five frequencies was used to obtain the values ​​of the slopes of the present embodiment, and ultimately the value of C, since it provides intermediate values ​​between the two extremes, ideal and no frequency jumping.

The procedure is terminated in step 299. Those skilled in the art will appreciate that steps 231 to 236 in Fig. 5 are basically equivalent to step 230 in Fig. 4.

Of course, a mobile station (MS), which estimates C / I incorrectly, is not the only situation in which a high FER level can be obtained. A high FER value in the lowest, most robust code mode is obtained when the radio conditions are poor, regardless of the threshold values, and therefore the FER values ​​obtained when in the lowest code mode should be discarded, at least if they are high. Accordingly, in one embodiment, one of the speech quality limits, the upper FER limit, can be set to infinity for the most robust code mode.

By analogy, a low FER value is obtained in the highest, least robust code mode when the radio conditions are excellent, regardless of the threshold values, and therefore the FER values ​​obtained when in the highest code mode should be discarded, at least if they are low. Accordingly, in one embodiment, one of the speech quality limits, the lower FER limit, can be set to infinity for the least robust code mode.

When the link channel ratio is so poor that a high FER level is obtained even if the MS uses the lowest code mode, further considerations must be made. In this case, of course, no threshold changes should be made. To distinguish between different cases with high FER levels, we would have liked to have had the relationship between the different codec modes used during the last EMR. A high EMR level and a high proportion of the lowest code mode would then indicate a poor channel and not an MS with an incorrect C / I determination. Unfortunately, this information is typically not available in the present embodiment. In the present embodiment, we must instead use what is available in the EMR, namely the latest code model used by the MS.

As illustrated in Fig. 6, a particular embodiment includes a step 233 which determines whether or not a reported FER value should be used for updating the FER array. For example, if the latest code mode used by the MS is the lowest code mode in the current ACS, the reported FER value can be discarded. Such a FER value will then not affect the long-term average. In order not to prevent low FER values ​​obtained in the lowest code mode from affecting FERit, it is preferable if the discard step also depends on the actual FER value, so that e.g. an FER value below the FER target value is allowed to be included in the calculation of FERit.

One analogous routine is to discard FER values ​​obtained in the highest code mode, especially when such FER values ​​are low.

Note that the present embodiment does not stop the entire algorithm when communication takes place via the lowest (or highest) code mode, but proceeds with the threshold adjustment calculations. Only the FER array update is not performed if the lowest (or highest) code mode is used. The reason for this is a little obscure and maybe a bit far-fetched, but there is a certain possibility that the mobile station (MS) has been in a different codec mode than the lowest one, and that the estimated FERit is higher than the upper FER limit, but that you are prevented. to adjust the thresholds for other reasons. Then suppose the radio quality drops so that a codec mode switch to the lowest codec mode occurs, and the adaptation restriction is lifted simultaneously. We now have the situation where we know that codec switching thresholds should be adjusted, but if we did not allow the example algorithm to continue when we are in the lowest mode, no adjustment would be made. The example algorithm would remain in this state until the radio channel gets better and we switch to a higher codec mode. The example custom threshold algorithm now allows you to adjust the threshold and does so, and the new modem switch threshold would probably cause an immediate down regulation to the lowest code mode, creating two codec mode switches (one upgrade followed by one downgrade) that closely follow. By contrast, if we allow the example threshold adjustment algorithm to adjust the thresholds even if we are in the lowest codec mode, we thereby avoid this unnecessary up and down codec mode.

A simulation according to the embodiment illustrated in Fig. 6 has been performed using an AMR link simulator. The incorrect estimation of C / I was simulated by adding a C / I offset to the link matching routine, just before the actual codec selection, causing the link matching to believe that the channel had a higher C / I than it actually had and, consequently, sometimes incorrectly chose a higher one. kodekmod. Four different variants were simulated, normal with a correct C / I estimate, + 2dB with C / I constant overestimated by 2 dB, + 5dB with C / I constant ε · Π '· 0 Ο&gt;) 7

£. V 4. * Ο overestimated by 5 dB and + 8dB with C / I constantly overestimated by 8 dB. The simulations were based on a version of the embodiment where the calculations of the threshold adjustments for AT2 and FR were replaced by simply setting the threshold adjustments to 2.5 dB.

The channel was TU3 and a frequency jump scenario was simulated, which skipped 5 frequencies (5FH). The length of the simulations was 22000 speech frames, ie. 440 seconds. The C / I profile was a continuously varying profile in which C / I varied between ~ 21 dB and ~ 3 dB.

Mean FERs from the simulations were dotted in Fig. 7. The FER values ​​given here are calculated only for frames in which the lowest code mode has not been used. Note that the FER values ​​are averages calculated over the full simulation run and this means that it includes frame deletions that occurred before the threshold matching algorithm had been stabilized.

From Fig. 7 it can be seen that both variants of the example threshold threshold algorithm work quite well and both manage to reduce FER quite substantially. In most cases, the AT2 variant exceeds ATI.

As mentioned further above, there are certain different approaches to how the thresholds should be updated. Fig. 8A illustrates an extreme approach when only the particular threshold around which the link quality is present is adjusted. Fig. 8B illustrates the second extreme approach, where all thresholds are adjusted by the same amount, regardless of the object of the evaluation. FIG. 8C illustrates another embodiment in which all the thresholds are adjusted, but where the amount of adjustment differs from one threshold to the other according to predetermined conditions. Fig. 8D illustrates a variant of Fig. 8A, where only one threshold is initially adjusted. However, in this embodiment, the adjustment procedure comprises an evaluation step, where conclusions are drawn if the adjusted threshold passes a different threshold. If this is the case, the threshold that is passed is also adjusted to keep the thresholds in the same order as they originally were.

Another aspect of the customization is if the customization is done individually, for a certain group of users or terminals, or in general. In one aspect, it is intended that the thresholds should be the same for each mobile station or user connected to the cell to which the present invention is applied. Such an approach will easily compensate for systematic errors in setting the original thresholds. Even errors made by the base station controlling the cell can be compensated in this way. However, errors made by the individual mobile stations cannot be generally compensated. Instead, such an approach will become somewhat unstable if terminals having substantially different C / I measurements aim in different directions.

If systematic mobile station errors are believed to depend on the actual manufacturer or mobile station model number, the threshold adjustments could be made valid for all mobile stations belonging to a particular group of mobile stations. Knowledge of the manufacturer or model number must then be available for the node performing the adaptation.

Individual mobile stations may also have slightly different systematic errors, which can only be compensated for by allowing each individual user to have their own set of thresholds. Fig. 8E illustrates such an approach. An initial threshold is adjusted for a first mobile station by an amount β, while the threshold is kept unchanged for a second mobile station.

Furthermore, the original or original set of thresholds used by a user in establishing a call can be selected in different ways. One way is to choose an original set of thresholds that is equal to the set most recently used by the user. This can be practically possible if the mobile station saves information on the thresholds used. However, since the number of potential users is huge, this approach is not very likely if only the base station system has

528 213 access to the actual thresholds. In such cases, a preferred method would instead be to base the original set of thresholds on a most recently used set of thresholds for the same type of receiver, e.g. the same manufacturer or model number.

This startup procedure could be improved by recording statistics on threshold adjustments and receiver characteristics. An original set of thresholds when making calls can then be made depending on the statistical behavior of recipients who have similar recipient characteristics.

As mentioned above, the present invention can be applied to both uplink and downlink communication. The present invention can also be applied to both the mobile station side and the base station system side. Figures 9A-D illustrate these alternatives.

In Fig. 9A, downlink communication 12 is considered. The mobile station 20 comprises measuring means 50 for measuring the speech quality in the received speech from the downlink

12th The mobile station 20 further comprises a code quality threshold adjustment unit 51 which performs the evaluation procedures of the present invention and provides suggestions for threshold adjustments. An inquiry 24 to make such adaptations is transmitted to a communication network node, in this case the base station 10, which decides whether the adaptations should be performed. Alternatively, if the system so permits, the mobile station 20 can perform the threshold adjustments.

In Fig. 9B, downlink communication 12. Again, the mobile station 20 also includes here measuring means 50 for measuring the speech quality in the received speech from the downlink 12. The mobile station 20 reports the result of the measurements uplinked to a communication network node, in this particular case the base station 10. In this embodiment, base station 10 comprises the code qualifier threshold matching unit 51 which performs the evaluation procedures of the present invention and the base station 10 performs

328 213 adaptation. The thresholds and / or the adjustment history of the thresholds are preferably stored in a memory 52.

In Fig. 9C, communication in the uplink 22 is taken into account instead. A communication network node, in this case the base station 10, here comprises the measuring means 50 for measuring the speech quality in the received speech from the uplink 22. The base station 10 also comprises in this embodiment the code quality threshold matching 51, which performs the evaluation procedures of the present invention. The base station 10 performs the adaptation and sends a command 14 to the mobile station about which codec mode to use.

In Fig. 9D, communication is considered in the uplink 22. This embodiment is probably the least useful. A communication network node, in this case the base station 10, comprises the measuring means 50 for measuring the speech quality in the received speech from the uplink 22. The base station 10 sends a report 15 on the result of the measurements to the mobile station 20. In this embodiment, the mobile station 20 includes the code qualification threshold adjustment 51, which performs the evaluation procedures of the present invention and provides threshold adjustment suggestions. An inquiry 24 for making such adjustments is transmitted to the base station 10, which decides whether the adaptations should be performed. Alternatively, if the system so permits, the mobile station 20 can perform the threshold adjustments.

In the various embodiments above, reports, commands and requests are transmitted between the base station and the mobile station. As described above, the EMR is a good candidate for transferring measurement results. However, both the voice quality measurement and commands and / or requests can be transmitted between recipients and transmitters using any communication channel, e.g. EMR, inband signaling, control channel signaling etc.

The present invention is intended to operate with AMR and AMR-like speech and audio codecs, e.g. AMR-WB and AMR-WB +, regardless of the particular r ··· '* —w

4 »jr the radio access method, but can also work with other code selection techniques that are based on link quality thresholds.

In the embodiments described above, the means for carrying out the methods of the invention is described as being included in the mobile terminal and / or the base station. However, the communication network means may also be included in communication network nodes other than the base station, e.g. in a base station controller or any other node connected to the base station. For example, in the embodiment of FIG. 9C, the measuring means 50 could be provided in the base station, while the code quality threshold matching unit 51 and / or memory 52 may instead be provided in a base station controller.

The embodiments described above are to be understood as some illustrative examples of the present invention. It will be appreciated by those skilled in the art that various modifications, combinations and modifications may be made to the embodiments without departing from the scope of the present invention. In particular, different sub-solutions in different embodiments can be combined to other configurations where technically possible. However, the scope of the present invention is defined by the appended claims.

REFERENCES [1] 3GPPTS 26.071, AMR speech codec; General description.

[2] 3GPPTS 26,171, Wideband AMR speech codec; General description.

[3] 3GPPTS 45.009: Link Adaptation [4] S. Wanstedt, J. Petterson, X. Tan, and G. Heikkila, Development of an objective speech quality measurement model for the AMR codec, MESAQIN 2002.

[5] ITU-T P.862, Perceptual evaluation of speech quality (PESQ).

[6] 3GPP TS 45.008, Radio subsystem link control.

528 213