You arrive home from a long day at work, flip on the classic stereo system you've spent nights and weekends restoring to perfection, and tune to an FM broadcast of a live local concert. As you settle back to enjoy it, you notice something peculiar. The normally pristine audio from a station that prides itself on a high-quality signal has an annoying background noise. You check your directional rooftop antenna, but it's aimed right at the station not more than ten miles away. Your tuner's signal-strength meter is pegged to the right, as usual. What gives? Then you vaguely recall that the station recently announced it would soon add HD Radio.
In its current hybrid implementation, HD Radio adds digital sidebands to an analog FM signal. The sidebands consist of hundreds of digital subcarriers that occupy spectrum 129 to 198 kHz from the channel center frequency. Total power for both sidebands originally was 20 dB below the analog carrier level, or −20 dBc. FCC rules now permit levels up to −14 dBc for most stations without special approval, and up to −10 dBc in certain circumstances. Some transmission modes use additional subcarriers that extend the sidebands as close as 102 kHz from channel center. These extended hybrid signals are becoming more common.
This is the RF spectrum of an HD Radio signal with digital sidebands at −20 dBc. The horizontal scale is 50 kHz per division and the vertical scale is 10 dB per division.
Although the digital sidebands occupy spectrum beyond the normal bandwidth of an analog signal, they remain outside the receive passband of other channels that may be locally allocated. But they can easily interfere with reception of a distant station on an adjacent channel 200 kHz away since the signal spectra overlap. FCC rules do not protect long-distance reception from interference.
An HD Radio station's digital sidebands may interfere with reception of its own analog signal in two ways. First, FM detection of the sidebands may generate intermodulation products that overlap the analog signal. For more about this problem, see Dave Hershberger's articles here and here. Interference also may occur when the stereo decoder demodulates the digital sidebands along with the stereo subchannel signal. Because the digital subcarriers are numerous and their data randomized, analog detection yields noise. This form of HD Radio self-noise is the subject of this article.
A stereo encoder lowpass-filters the left and right audio channels to 15 kHz and adds them to form the L+R signal. It subtracts them to get L−R and amplitude-modulates this signal onto a 38-kHz subcarrier. The encoder suppresses the subcarrier and replaces it with a phase-coherent 19-kHz pilot. It combines these signals to form a composite signal consisting of L+R to 15 kHz, pilot at 19 kHz, and double-sided L−R from 23 to 53 kHz. The composite signal frequency-modulates the RF carrier.
To recover the left and right channels, a receiver detects the FM signal and passes the composite signal to the stereo decoder. The decoder phase-locks a 38-kHz oscillator to the 19-kHz pilot, demodulates the L−R signal to baseband with the oscillator, and then adds and subtracts the baseband L+R and L−R signals to get L and R. Lowpass filters eliminate everything above 15 kHz.
An ideal way to demodulate the L−R signal is with a multiplier. Multiplying the composite signal by a 38-kHz sine wave phase-locked to the 19-kHz pilot yields L−R at baseband plus ultrasonic mixing products. Lowpass filtering leaves L−R. This demodulator has no inherent spurious responses. Practical issues with four-quadrant analog multipliers include linearity, temperature stability, complexity, and cost. A few high-end tuners have used this approach in discrete stereo decoders, but it seems never to have been implemented as a standard integrated circuit (a custom Delco chip is described here).
IC stereo decoders instead use synchronous signal inversion. The composite signal is passed either unaltered or inverted, with changeover at a 76-kHz rate synchronized to the 19-kHz pilot. The signal never undergoes variable attenuation as with an analog multiplier, so linearity depends almost entirely on the fidelity of the inversion transitions. This is easily controlled in integrated circuitry. This approach effectively multiplies the composite signal by a 38-kHz square wave with amplitude levels of +1 and −1. A square wave contains odd harmonics of the fundamental frequency, in particular the third at 114 kHz and the fifth at 190 kHz. But since the composite signal has no components near these frequencies, it decodes as if a sinewave multiplier had been used.
The problem with squarewave decoding is that detected HD Radio signals have power near 190 kHz, and extended hybrid signals have additional power near 114 kHz. The squarewave harmonics demodulate this power to baseband. This is the source of directly demodulated HD Radio self-noise.
The four-quadrant analog multiplier is a special stereo decoder. So is a harmonic canceller. This type of decoder uses a switched, multilevel waveform to eliminate one or more squarewave harmonics without resorting to an analog multiplier.
This four-level Walsh-function waveform, generated by a discrete Sansui stereo decoder, has no third or fifth harmonic. The first nonzero harmonic at 266 kHz is beyond the spectrum of a detected HD Radio signal.
Pioneer Electronics developed a novel stereo decoder that uses two custom ICs and includes a pulse-count FM detector. Reversing roles, the binary composite signal passes or inverts a 38-kHz sine wave locked to the 19-kHz pilot. Since the pulse rate is above 1 MHz, spurious responses lie well beyond the spectrum of a detected HD Radio signal. (The multiplication circuits are analog switches, not four-quadrant multipliers.)
Finally, the composite signal can be decoded with digital signal processing. The 100-dB stopband rejection typical of audio A/D anti-alias filters removes detected HD Radio signals, while L−R demodulation by numerical multiplication incurs no linearity penalty.
The HD Radio self-noise level depends on the digital sideband level and HD Radio service mode, and on the receiver IF bandwidth, postdetection filter, and stereo decoder. Home tuners and receivers with two wideband ceramic IF filters, no postdetection filter, and a squarewave stereo decoder IC are common. For such systems HD Radio self-noise typically is about 50 dB below the level of a 100%-modulated, 1-kHz sine wave. The derivation of this figure is here. The noise level for most home equipment made in the past 35 years should be within a few dB of this figure. It can be up to 10 dB higher for an extended hybrid HD Radio signal. It will increase n dB when the digital sidebands are n dB above −20 dBc. FCC rules now permit −14 dBc for any station and −10 dBc with special permission.
I've measured the RMS level of FM program audio as anywhere from 5 to roughly 25 dB below that of a 100%-modulated, 1-kHz sine wave. The first figure is for highly compressed pop or rock, while the second is for quiet passages in classical music. HD Radio self-noise on home equipment therefore is just 25 to 45 dB below the program level. This explains why I've found HD Radio signals unlistenable on every home tuner or receiver I've tried that lacked a special stereo decoder, narrow IF filter, or postdetection filter.
Although some equipment is more elaborate, home tuners and receivers commonly have used a cascaded pair of 280-kHz ceramic IF filters to set the receive passband.
This shows the detected spectrum in one such tuner when receiving an HD Radio signal. The horizontal span is 200 kHz and the vertical scale is 10 dB per division. The program material was monophonic. The signals to the left of midscreen are L+R, pilot, residual L−R near −30 dB, RDS, and two SCAs. To the right is the HD Radio signal.
The HD Radio spectrum, flat at RF, rises 6 dB per octave after FM detection. The roll-off of cascaded 280-kHz IF filters matches the rise around 150 kHz, and the filters take over above. But the spectrum is not down much between 175 and 198 kHz, the region a squarewave stereo decoder will demodulate to baseband with its 190-kHz fifth harmonic.
Narrower IF filters make quite a difference. This shows overlapped spectra for three different sets of IF filters in the same tuner. The traces are for pairs of 280-, 180-, and 150-kHz filters. The differences in HD Radio spectral level are pronounced, as are the differences in audible self-noise.
Narrow IF filters increase harmonic distortion, exaggerate multipath distortion, decrease modulation acceptance, and degrade stereo separation. Elevated multipath distortion can be quite audible. Less noticeable are ticks and pops coincident with audio peaks on stations that overdeviate. Higher harmonic distortion and lower stereo separation are measurable but not necessarily audible. Some equipment includes compensation to restore separation in narrow. Manufacturers almost always use wide filters in home equipment with a single IF bandwidth. This makes HD Radio self-noise louder than it otherwise might be.
I've never heard HD Radio self-noise in a car radio. This may be due to the narrower IF filters commonly employed or to more complex signal processing. Reception is more difficult in a moving vehicle and more people listen on the road than at home, so advanced techniques such as digital signal processing or postdetection filtering are more common in car radio designs.
Few home tuners or receivers made for the U.S. market provide a dedicated lowpass filter between the FM detector and stereo decoder. Most just use a single capacitor to suppress IF ripple. Audio bandwidth and stereo separation depend on the amplitude and phase response of the 53-kHz composite-signal passband. A postdetection filter that makes a significant dent below 200 kHz needs careful design to maintain flat composite response. Nevertheless, some high-end tuners do use such filters, often multipole/multizero designs with response nulls at 114 and 190 kHz, the third and fifth harmonics of 38 kHz.
This passive postdetection filter in a Kenwood KT-880D tuner completely eliminates HD Radio self-noise.
Postdetection filters sometimes are called birdie or anti-birdie filters when intended to prevent the demodulation of extraneous composite signals as tones. Such filters may or may not be effective against HD Radio self-noise. While some are complex lowpass filters, others are just simple LC traps resonant at the 67-kHz SCA frequency.
The European version of a tuner often includes an LC postdetection filter missing in the U.S. version. The filter reduces interference from adjacent-channel signals that may be only 100 kHz from the tuned frequency. These filters should greatly reduce or eliminate HD Radio self-noise.
Mono The easiest way to get rid of HD Radio self-noise is to hit the mono button. The noise corrupts only the stereo subchannel. If you make no use of the L−R signal, no noise will appear. I don't like listening to stereo music monophonically, but I often listen to monophonic spoken programs on public radio stations. Some stations kill the stereo pilot for these programs, causing your receiver to revert to mono and any HD Radio self-noise to disappear. For stations that retain the pilot, press the mono button to eliminate HD Radio self-noise.
Hi-Blend If a signal has stereo content and you'd like to preserve some of it, engage hi-blend. This filter rolls off the highs of the L−R signal before combining it with L+R, typically providing 3–6 dB of noise reduction and 10–20 dB of midrange stereo separation. Hi-blend noticeably quiets the audio while retaining a surprisingly good stereo effect. However, HD Radio self-noise will remain audible unless the listening level is low. (The hi-blend function may be called subchannel filter or multiplex filter. But in some equipment, what's called a multiplex filter lowpasses L and R, not L−R, and will not reduce HD Radio self-noise.)
Narrow IF Filter Obtaining a wide stereo image without HD Radio self-noise is difficult. If your tuner has a selectable narrow IF bandwidth, try it. Even if this merely cascades additional wideband filters to improve alternate-channel selectivity, HD Radio self-noise should noticeably drop. Filter bandwidths of 180 kHz or less will drop the noise substantially. Using such a filter may not eliminate the last traces of HD Radio self-noise, but I think the result will satisfy most listeners.
Although narrow filters measurably degrade signal fidelity, I've never been able to hear a difference in sound quality between wide and narrow IF filters on speech or music, except when receiving multipath-laden or overdeviated signals. You may occasionally object to the sound of a narrow IF filter, but in most cases I think you'll find the reduction in HD Radio self-noise worthwhile.
You can retrofit your own equipment with narrower ceramic filters. Digi-Key and Mouser stock a variety of bandwidths, and filters sometimes are available on eBay. Filter characterization and tuner realignment are necessary for lowest distortion and best performance. But if you obtain a dozen filters and a breakaway in-line socket strip with which to make three-pin filter sockets, you can swap filters until you're satisfied with the sound.
Postdetection Filter A solution with negligible sonic penalty is to add a postdetection filter. You can make one with an op-amp and a few resistors and capacitors.
This particular design provides 50 dB of stereo separation and a frequency response flat within 0.5 dB. Response is down 45 dB at 190 kHz, which renders hybrid HD Radio self-noise inaudible. It is down 31 dB at 114 kHz, which may leave just a trace of residual noise on an extended hybrid signal. You can design a filter to your own specifications with software you can download here, where the filter and its application are more fully explained.
This shows a postdetection filter installed in a Carver TX-11b tuner.
Tuners The last solution is simply to buy another tuner. Home tuners are out of fashion these days and few new ones are marketed. But high-quality used tuners are readily available.
Any tuner with a selectable narrow IF bandwidth will reduce HD Radio self-noise, perhaps to below audibility. You'll have to put up with any degradation in harmonic distortion, multipath rejection, modulation acceptance, or stereo separation that results. Tuners that cascade more than two wideband IF filters will reduce HD Radio self-noise, but not to inaudibility. The effect of IF filters is more fully explored here.
To receive an HD Radio signal without self-noise in stereo at full IF bandwidth, a postdetection filter or special stereo decoder is essential. The tuners listed below have one or the other. (If a tuner has international versions, I list it only if the U.S. version qualifies. I have not listed any HD Radio tuners. They use DSP for FM detection and stereo decoding and should be free of HD Radio self-noise during analog reception.) I have verified the effectiveness of self-noise rejection only for a few of the tuners listed. Surely additional tuners qualify for the list.
Postdetection Filter ADS/Braun T2 AMC T7 Carver TX-11 Denon TU-767, TU-800 Harman Kardon TU920 Hitachi FT-5500MKII, FT-8000 Kenwood 600T, KT-80, KT-880D, KT-917, KT-5020, L-07TII Klein + Hummel FM 2002 Marantz 2130 McIntosh MR71, MR73, MR74, MR77, MR78 Onkyo T-4087, T-4150, T-4500, T-4700, T-4711, T-9060, T-9090, T-9090II, T-G10 Phase Linear T 5200 Pioneer F-9, TX-8500II, TX-9100 Sansui TU-717, TU-919, TU-X1 Sony ST-J75, ST-S555ES Sumo Charlie Teac TX-500 Technics ST-G7, ST-G70, ST-S8 Analog Multiplier Decoder Kenwood KT-990D, KT-1100SD, KT-3300D, KT-7020, L-1000T Rotel RHT-10, RT-990BX Yamaha T-85, TX-1000, TX-2000 Walsh Function Decoder Denon TU-680NAB Sansui TU-S77X, TU-S77AMX, TU-D99X, TU-D99AMX Sony ST-S800ES Switched-Sinewave Decoder Pioneer F-77, F-90, F-91, F-93, F-99X, F-449, F-717 DSP Decoder Accuphase T-1000, T-1100
I recorded this sound sample from a Proton 300 table radio. It uses three 280-kHz IF filters, no postdetection filter, and a squarewave stereo decoder. The RF signal level was 73 dBf, the digital sideband level was −14 dBc, and the HD Radio service mode was MP3. The RMS level of the unmodulated section is −47.6 dBfs and that of the voice portion is −19.5 dBfs. Therefore S/N is 28.1 dB.