Nearly all stereo decoders use a 38-kHz square wave for demodulating the L−R subchannel, which lies between 23 and 53 kHz. An undesired consequence is that the waveform's fifth harmonic demodulates power near 190 kHz. HD Radio digital sidebands, which occupy spectrum from 129 to 198 kHz after FM detection, can cause an annoying audio background noise when demodulated by the fifth harmonic. Extended hybrid HD Radio signals, whose detected spectrum may extend as low as 102 kHz, can cause additional noise when demodulated by the third harmonic at 114 kHz. Adding a lowpass filter between the detector and stereo decoder can eliminate this HD Radio self-noise.
To avoid degrading frequency response and stereo separation, postdetection filters typically exhibit flat amplitude and group delay over the 53-kHz stereo-composite passband. Although it rolls off slowly, active filters usually employ a Bessel response. For example, the Philips TEA6880H stereo-decoder chip for car radios includes an eight-pole, 80-kHz Bessel filter that provides 19 dB of attenuation at 190 kHz. The TDA1502 yields 17 dB with a four-pole filter. With the help of narrow IF filters and a little road noise, these rather anemic postdetection filters still may render HD Radio self-noise inaudible.
To address the problem for home tuners with wide IF filters in a quiet environment, I wrote a computer program to model the stereo decoding process and optimize a postdetection filter. The program seeks the filter with maximum attenuation at 190 kHz for a specified minimum stereo separation and maximum frequency response error. It imposes no spectral constraint on the filter itself and directly optimizes the pole locations instead of using a classical response function. By including the stereo decoder in the model, the optimizer can take advantage of redundancy in the double-sideband L−R signal. This approach requires just three poles.
The program couples a downhill-simplex local optimizer with a stochastic global optimizer. The resulting optimal filter is down 3 dB somewhere between 30 and 50 kHz. Roll-off this low attenuates the L−R signal, particularly the upper sideband. But like the vestigial sideband system of NTSC television, the vector sum of the demodulated sidebands is nearly constant. When installing the filter, you compensate for the additional L−R attenuation by readjusting the tuner's stereo separation control. (Often you must change a fixed resistor that limits the control range.)
Freq Sep L R L+R L-R 1000 68 0.00 -68.36 0.00 0.00 2000 62 -0.01 -62.25 -0.01 -0.01 3000 59 -0.02 -58.61 -0.02 -0.02 4000 56 -0.04 -56.00 -0.04 -0.03 5000 54 -0.06 -54.02 -0.06 -0.05 6000 52 -0.08 -52.51 -0.09 -0.08 7000 51 -0.11 -51.40 -0.12 -0.11 8000 50 -0.15 -50.64 -0.16 -0.14 9000 50 -0.19 -50.25 -0.20 -0.18 10000 50 -0.23 -50.23 -0.24 -0.22 11000 50 -0.27 -50.59 -0.29 -0.26 12000 51 -0.32 -51.31 -0.34 -0.30 13000 52 -0.37 -52.13 -0.39 -0.35 14000 52 -0.42 -52.17 -0.44 -0.40 15000 50 -0.47 -50.47 -0.50 -0.45 Fc = 34864 Hz Inj = -4.0 dB 3rd = -30.2 dB 5th = -43.7 dB S/N = -0.0 dB to 5 kHz, +0.1 dB to 10 kHz, +0.2 dB to 15 kHz R = 2400 Ω C1 = 6250 pF C2 = 2913 pF C3 = 361 pF
This is the filter optimizer output for 50-dB minimum stereo separation and 0.5-dB maximum frequency response error. Sep is stereo separation in dB. L and R are the demodulated levels for a left-only signal. L+R and L−R are the levels for monophonic and antiphase signals. All figures assume perfect performance before adding the filter.
Fc is the filter −3-dB corner frequency. 3rd and 5th are the attenuations at the third and fifth harmonics of the 38-kHz stereo-decoder oscillator where the HD Radio noise lives. See S/N Enhancement for an explanation of S/N.
Inj is the required L+R stereo matrix injection. The audio level will change by this amount or less when you install the filter and readjust 1-kHz separation for maximum. Stock L+R injection often is about -1 dB so the actual level change may be somewhat less than indicated. Usually the lower output isn't a problem, but you can restore it to normal by increasing the gain of the detector, stereo decoder op-amps, or output amplifiers.
These are the filter amplitude, phase, and group-delay responses from a circuit simulation program.
This is the filter circuit. Any wideband, low-distortion op-amp will work. Add a 0.1-µF ceramic between +V and −V if the supplies aren't bypassed to ground nearby. Ground the −V terminal in a single-supply system. If the detector output impedance isn't low, use a TL072 and configure the second op-amp as a voltage follower. I use selected parts within 1% of the values shown, but 5% parts should be good enough. You can specify the filter resistance in the optimizer.
When calculating stereo separation, the optimizer assumes that your tuner compensates for any excess phase shift between 19 and 38 kHz due to the IF filter or FM detector. This may not be the case unless your stereo decoder has a capacitor like C1 in the Sanyo LA3450 circuit or C212 in the Hitachi HA11223W circuit shown above. For best separation you may need to add such a capacitor. For the AN363, AN7470, HA1156W, HA1196, LA3400, LA3401, LA3410, and µPC1235C it goes on pin 3; for the µPC1223C, pin 18; and for the TCA4500A, pin 2. The capacitor may affect stereo distortion for some decoders.
This shows convergence zones in the complex plane for the global optimizer. Blue and red dots are local-optimizer starting points, while yellow and cyan are where it wound up. The black region is beyond the filter specifications and remains unexplored. The global optimum is at the rightmost cusp.
A postdetection filter can reduce noise for any stereo signal, not just one with HD Radio sidebands.
Detected FM noise increases 6 dB per octave, the same rate that squarewave harmonic amplitudes decrease. Thus each 38-kHz harmonic can potentially contribute as much noise as that in the L−R region. The IF filter will attenuate some of this harmonic noise. A postdetection filter can eliminate the rest.
There is one more noise effect. Because detected FM noise increases with frequency, the upper L−R sideband is noisier than the lower sideband. A postdetection filter may increase S/N by reducing the USB noise contribution. It may decrease it by lowering the signal level even more, since signal components combine coherently while sideband noise combines incoherently. The net S/N effect depends on the particular filter response. The S/N figures in the filter optimizer output show the small gain or loss over three frequency bands.
For the wide IF filter (two 250-kHz Murata MXs), 50-dB stereo quieting sensitivity for a Yamaha T-1020 was 42.4 dBf. Adding the postdetection filter increased sensitivity 2.4 dB to 40.0 dBf. For the narrow IF filter (two 110-kHz Murata MHYs cascaded with the 250s), sensitivity increased 1.5 dB from 40.6 dBf to 39.1 dBf with the postdetection filter.
This is the detected spectrum to 200 kHz for an HD Radio signal in a Yamaha T-1020 (wide IF filter).
This is the spectrum after installing the postdetection filter.
This shows the filter installed on a perfboard in the tuner. The T-1020 uses a noise-detection bandpass filter in the 125-kHz region to automatically select the IF bandwidth and the stereo/mono mode. After installing the postdetection filter, the tuner still thought most clean signals were noisy. I had to route the postdetection filter output to the noise filter and then boost its gain somewhat to restore normal operation. This is typical of the complications you may encounter when adding a postdetection filter.
Here a postdetection filter is installed in a Sony ST-S444ESX. This filter uses 2.7kΩ resistors. I selected three that measured within a few ohms of 2811Ω and then used this value in the filter optimizer to determine the capacitor values. Adding a 750-pF phase-compensation capacitor to the CXA1064 stereo decoder increased 1-kHz stereo separation from the high-40s to the mid-60s in dB.
This shows a postdetection filter built directly on the PCB in a Technics ST-9030 tuner. Parallel capacitors comprise C1 and C2. The 1kΩ resistor is not part of the filter circuit.
Harmonic cancellers, which are implemented as nonlinear postdetection filters, are described here.
88–108 MHz