I have been reading the discussion about reducing AM audio frequency response to 5 kHz. First Clear Channel implemented it, then Crawford. Now even Radio World has endorsed it. In this editorial I provide a dissenting opinion and back it up with facts, not listening trips from the airport.
Right now AM bandwidth is fixed at 20 Hz-10 kHz. This is the result of the NRSC standards that were set over a decade ago. Prior to this standard, AM stations had no limit on their high-frequency response, which caused problems due to the 10 kHz spacing between AM channels. Add to that the extreme high-frequency pre-emphasis employed by many stations at the time, and you had a recipe for splatter, with some stations 30 kHz (or more) wide.
The NRSC studied the problem and came out with a sane answer: a 10 kHz steep audio roll-off.
Group delay distortion
The 10 kHz bandwidth reduction is achieved through a steep-slope low-pass audio filter just before the transmitter that cuts off everything above 10 kHz. One look at a spectrum analyzer display of an AM station is worth 1,000 words; the sidebands literally fall off a cliff above 10 kHz.
Unfortunately, steep-slope filters can cause audible artifacts. Their biggest problem is called group delay distortion. To understand what this is, let me explain.
We all know that playing an "A" on a piano sounds drastically different then setting an audio oscillator to 440 kHz. What makes them different is that the piano not only produces the fundamental frequency, it also has harmonics added. These harmonics are even or odd multiples of the fundamental frequency (880, 1320, 1760, etc.).
The harmonics extend past the limits of human hearing, which is 20 kHz. Each and every sound has its own unique harmonic structure. To recreate any sound accurately, the reproduction equipment must have flat frequency response, low distortion and noise and a flat time response. In other words, the entire audio waveform must arrive at your ear clearly, at the right level and in the proper time.
Group delay is exactly what it says: delay. As you approach the cut off frequency of a filter, the frequencies begin literally to slow down as they go through the filter. This means that they arrive after the fundamental and other harmonics. Problem is, humans can hear time delay distortions and filter group delay quite easily.
We usually perceive group delay as a "phasiness" to the audio. Group delay gets worse with the complexity of the filter. This is why audio purists prefer gentle-slope filters to steep-slope ones in crossovers. It's also the reason the old Dorrough DAP sounded so good. Fortunately, the NRSC was aware of this problem and set the cutoff frequency of the NRSC filter at 10 kHz, meaning that only frequencies above 7 kHz would experience severe group delay.
Another characteristic of human hearing is that it is not flat - the average adult has a broad peak in his hearing response centered at 3 kHz. This is why a small adjustment of the 3 kHz band on a graphic equalizer makes such a big difference compared to the high- and low-frequency bands.
Now let's see what happens when we move the frequency response of that NRSC 10 kHz audio filter down to 5 kHz. First the bandwidth reduces to 5 kHz. Next, group delay distortions are now occurring at 3.5 kHz. In other words, we have introduced a time distortion right where our hearing is the most sensitive! Most of us know that some forms of distortions multiply rather then add; indeed, that is the case here.
Some engineers with good intentions have replied to me saying modern computer digital filter design can produce audio filters that have low group delay response right to their cutoff. This is quite true. But there's a second filter that is in cascade with the audio filter over which we have no control: the filter in the receiver.
It's well known and documented that most AM radios have terrible high-frequency response. Manufacturers do that because it is a cheap way of improving selectivity in the radio. AM stations have been employing pre-emphasis (high-frequency boost) for decades to fight it. The NRSC even publishes a desired pre-emphasis curve.
We've all seen the frequency response measurements of a "typical" AM radio - audio response down 6 or more dB at 4 kHz. Let me clue you into something else: the IF filters in the radio also produce group delay. This group delay can multiply with the delay in the audio filter causing severe audio artifacts. The NRSC realized this when they set the cutoff frequency of their filter above the response of the typical AM radio. Now some want to put it right in the middle of the typical AM radio's response.
This is a disaster waiting to happen.
If you don't believe me, listen to any AM talk station that's implemented these filters. Here in Los Angeles it's KFI(AM), and their phone calls sound far better then their announcers do. Why? Because the phone is steep-slope filtered at 3 kHz, well below the frequency response of the audio filter. The announcers have this "phasey" sound that's not on the calls.
In the early 1990s I was chief engineer at WHDH(AM) 850, Boston, one of the two big talk stations there (now known as WEEI). Our directional pattern was designed in the late 1940s and protected KOA in Denver. Unfortunately, the metro had grown west quite a bit, and listeners west of the transmitter in Needham used to complain they could not get us after sunset. We thankfully discovered a CRL audio processor designed for shortwave use, which made us loud.
One feature of the unit enabled the user to change the response of its audio filter. Over a period of several months, we experimented with audio responses of 4 kHz, 5 kHz, 7.5 kHz and 10 kHz. The entire radio and TV station was involved; even interns listened on air and gave their input.
The results were that audio responses of 4 and 5 kHz caused a "phasey" sound that most did not like. The effect was minimal at 7.5 kHz and could not be heard at all at 10 kHz. We settled on a final audio response of 7.5 kHz.
To me, this confirmed in the field what's predicted in the lab: Steep-slope audio filters with cutoffs at or below the typical response of an AM receiver should be avoided. Yet this is exactly what the proponents of limiting bandwidth to 5 kHz propose to do.
If you truly need the extra quarter dB of loudness, a good compromise is to employ a cutoff frequency of 7.5 kHz.
Over 20 years ago, I remember reading a sentence in the back of the original AM Optimod manual. It said, "The future belongs to the quality conscious." To me at least, that's even more valid today. Let's keep AM sounding good. Let's set AM technical policy based on solid engineering study and measurements, not "junk science."