CSI estimation and LLR approximation for QAM demodulation in FM HD radio receivers

1. A radio receiver comprising:

physical layer circuitry configured to receive quadrature amplitude modulation (QAM) symbols via a plurality of subcarriers included in a broadcast radio signal, wherein each received QAM symbol is a complex symbol comprising multiple bits of encoded source information;

processing circuitry configured to:

demodulate the received data symbols;

generate a constellation sample for each received QAM symbol;

generate a soft metric for each bit of encoded information in the received QAM symbols using the constellation sample;

select, according to a QAM modulation type of the received QAM symbol, a soft QAM metric nonlinearity to modify the value of the soft metric for each bit component of the constellation sample; and

multiply the modified soft metric by a channel state information (CSI) weight to produce a Log-likelihood Ratio (LLR) approximation for each bit of encoded information in the received QAM symbols.

2. The radio receiver of claim 1 , wherein the processing circuitry is configured to multiply the soft metric by a CSI weight that includes a bias-corrected ratio of a determined magnitude-squared channel gain and a determined symbol noise variance.

3. The radio receiver of claim 2 , wherein the bias-corrected ratio includes a symbol noise variance bias-corrected for error introduced by CSI estimation.

4. The radio receiver of claim 3 , wherein the bias-corrected ratio includes a symbol noise variance bias-corrected for error due to determining the symbol noise variance using a median-filtered estimate of the symbol noise variance.

5. The radio receiver of claim 3 , wherein the processing circuitry is configured to perform CSI noise variance estimation using an excess noise feedforward path when an excess noise level is present in the QAM symbol sample, wherein the excess noise feedforward path scales the noise variance estimate by an excess noise feedforward path gain; and

wherein the bias-corrected ratio includes a symbol noise variance bias-corrected for a bias error introduced by the excess noise feedforward path.

6. The radio receiver of claim 2 , wherein the bias-corrected ratio includes an estimated magnitude-squared channel gain bias-corrected for error introduced by CSI estimation.

7. The radio receiver of claim 1 , wherein the processing circuitry is configured to determine a complex soft metric of a received QAM symbol by modifying a constellation sample of a received QAM symbol using the soft QAM metric nonlinearity, wherein the soft QAM metric nonlinearity limits the value of the soft metric for each bit component of the received QAM symbol.

8. The radio receiver of claim 7 , wherein the soft QAM metric nonlinearity reduces the value of the soft metric for each bit component of the received QAM symbol.

9. The radio receiver of claim 7 , wherein the processing circuitry is configured to:

select a first soft QAM metric nonlinearity to modify the value of the soft metric for each bit component of the constellation sample when the QAM modulation includes four point QAM (4-QAM);

select a second soft QAM metric nonlinearity to modify the value of the soft metric for each bit component of the constellation sample when the QAM modulation includes sixteen point QAM (16-QAM); and

select a third soft QAM metric nonlinearity to modify the value of the soft metric for each bit component of the constellation sample when the QAM modulation includes sixty-four point QAM (64-QAM).

10. A radio receiver comprising:

physical layer circuitry configured to receive quadrature amplitude modulation (QAM) symbols via a plurality of subcarriers included in a broadcast radio signal, wherein each received QAM symbol is a complex symbol comprising multiple bits of encoded source information;

processing circuitry configured to:

generate a soft metric for each bit of encoded information in the received QAM symbols using a constellation sample generated for each received QAM symbol;

select a soft QAM metric nonlinearity to modify the value of the soft metric for each bit component of the constellation sample;

determine Log-likelihood Ratio (LLR) metric values for the received QAM symbols using modified values of the generated soft metrics;

scale the LLR metric values using an LLR scaling gain to produce scaled LLR metric values, wherein the LLR scaling gain is a function of a number of bits of quantization for a binary LLR value of the determined LLR metric values;

quantize the scaled LLR metric values to produce quantized scaled LLR metric values, wherein a step size of the quantization is a function of the number of bits of quantization; and

decode a received QAM symbol using the quantized scaled LLR metric values.

11. The radio receiver of claim 10 , wherein the processing circuitry is configured to:

multiply the soft metric by a channel state information (CSI) weight to produce an LLR metric value for each bit of encoded information in the received QAM symbols.

12. A method of generating Log-likelihood Ratio (LLR) approximations from a broadcast radio signal using channel state information (CSI) estimates, the method comprising:

receiving quadrature amplitude modulation (QAM) symbols via subcarriers of the broadcast radio signal and demodulating the QAM symbols, wherein each received QAM symbol is a complex symbol comprising multiple bits of encoded source information;

generating a constellation sample for each received QAM symbol;

generating a soft metric for each bit of encoded information in the received QAM symbols using the constellation sample;

selecting, according to a QAM modulation type of the received QAM symbol, a soft QAM metric nonlinearity to modify the value of the soft metric for each bit component of the constellation sample; and

multiplying the modified soft metric by a CSI weight to produce an LLR approximation for each bit of encoded information in the received QAM symbols.

13. The method of claim 12 , wherein multiplying the soft metric by the CSI weight includes multiplying the soft metric by a CSI weight that includes a bias-corrected ratio of a determined magnitude-squared channel gain and a determined symbol noise variance.

14. The method of claim 13 , wherein the bias-corrected ratio includes a symbol noise variance bias-corrected for error introduced by CSI estimation.

15. The method of claim 14 , wherein the bias-corrected ratio includes a symbol noise variance bias-corrected for error due to determining the symbol noise variance using a median-filtered estimate of the symbol noise variance.

16. The method of claim 14 ,

wherein the CSI estimation includes estimating the noise variance of a QAM symbol sample using excess noise feedforward path processing when an excess noise level is present in the QAM symbol sample, wherein the excess noise feedforward path scales the estimated noise variance by an excess noise feedforward path gain; and

wherein the bias-corrected ratio includes a symbol noise variance bias-corrected for a bias error introduced by the excess noise feedforward path processing.

17. The method of claim 13 , wherein the bias-corrected ratio includes an estimated magnitude-squared channel gain bias-corrected for error introduced by CSI estimation.

18. The method of claim 12 , wherein generating the soft metric for each received QAM symbol includes modifying the constellation sample of a received QAM symbol using the soft QAM metric nonlinearity, wherein the soft QAM metric nonlinearity is specified according to the QAM modulation type of the received QAM symbol, wherein the soft QAM metric nonlinearity limits the value of the soft metric for each bit component of the received QAM symbol.

19. The method of claim 18 , wherein modifying a constellation sample of a received QAM symbol includes:

selecting a first soft QAM metric nonlinear function to modify the value of the soft metric for each bit component of the constellation sample when channel noise includes more Gaussian noise than impulsive noise; and

selecting a second soft QAM metric nonlinear function to modify the value of the soft metric for each bit component of the constellation sample when the channel noise includes more impulsive noise than Gaussian noise.

20. The method of claim 12 , further including:

scaling a produced LLR approximation using an LLR scaling gain to produce a scaled LLR metric value, wherein the scaling gain is a function of a number of bits of quantization for a binary LLR value of the produced LLR approximation;

quantizing the scaled LLR metric value to produce a quantized scaled LLR metric value, wherein a step size of the quantization is a function of the number of bits of quantization; and

decoding a received QAM symbol using the quantized scaled LLR metric value.