AM bandwidth confusion: IF Filters today vs. yesterday
Whether or not a radio is a joy to listen to, or apt for difficult DX is decided to a large part by the last IF filter stage. Most radios of the past didn’t come with many IF filters to begin with, their quality varied a lot and harsh compromises had to be made in multi-purpose radios. With the advent of digital filtering and inexpensive DSP portables, we got spoiled by a rich choice of selectivity settings that came with an idiosyncrasy I had to wrap my head around first, namely a shift in how some (digital) filters are labeled now.
In AM mode both sidebands contain the same modulation information, meaning the audio bandwidth fitting into a 9kHz (region 1 and 3) or 10kHz (region 2) medium wave channel is approx. 1/2 of the modulation bandwidth – that’s 4.5 or 5 kHz. On shortwave, the 5kHz channel spacing is good for only 2.5kHz of audio spectrum.
Physical IF filters are specified by their entire (IF-) passband, not the audio passband. To enjoy the full audio spectrum an AM station can (usually) transmit, you need a 10kHz wide filter on the US medium wave (9kHz in the rest of the world), or a 5 kHz filter on shortwave. The more seasoned SWLs and hams among us are very used to that scheme, hence the prospect of a digital filter offering a maximum of only “6kHz” bandwidth or even less appears disappointing, even though that was the standard “wide” filter on many SWL tabletops and portables of the past..
When I got my first typical (Silicon Labs 473x-) DSP portable I started wondering about the filter choices: The widest filter setting showed familiar “6 kHz” in the display but it sounded so much wider than that! Checking the audio spectrum, I could see that the “6kHz” filter had full 6 kHz of audio bandwidth, with the other filter settings following that scheme. In other words, the settings were named after the audio bandwidth or ‘per-sideband’ bandwidth and didn’t follow the customary analog filter descriptions, the “6kHz” filter setting is in fact equivalent to a 12kHz IF filter in the old spelling!
Likewise, the Belka (ultra-) ultraportables – although not built around a SiLabs DSP – use the “audio bandwidth” naming scheme too, with the widest filter being called “4kHz”. This would be the equivalent to an oldschool 8kHz IF filter and it’s actually even wider – but this naming convention occasionally leads to unnecessary concerns about “too narrow” filter choices in these radios.
On hardware radios the naming shift might be just a means of simplification, as far as I know restricted to SiLabs-based DSP portables and the Belka, but while some SDR softwares (SDR Sharp, SDR Uno) are using the conventional ‘IF bandwidth’ naming scheme, SDR Console is using ‘AF/per sideband’ again. No surprise that this is confusing – but luckily only in the double sideband modulation modes.
In SSB only one sideband is supposed to go through the filter at a time, that’s why the filter’s specified bandwidth is equal to the possible audio bandwidth anyway and the befuddling naming issue doesn’t exist.
Demonstration and practical (kind of) measurements on some radios
While trying to make screenshots to illustrate the issue, I had the idea to generally check the filters of my radios for their characteristics. In order to show that I needed to cheat a little: Rather than trying to measure filters I can’t access individually for analysis with equipment I don’t own, I merely analyzed the audio output of the radios. While this can show an approximation to the actual filter shape, this is including every signal change happening behind the filters, for example tone controls. Lacking a signal generator, I tuned the radios to a 10kHz wide DRM signal from Kuwait on 15110 kHz to give the radios a wide, actual signal and circumvent any level-based automatic bandwidth or noise reduction shenanigans the DSP radios might employ.
The PL-660 is unfortunately my only example of an (almost) oldschool analog portable, sporting 2 conventional, inexpensive ceramic filters of unknown specification and not using DSP for that. Their model numbers remind a bit of the MuRata filter codes but that doesn’t correspond with the results shown below, the “CXE455IU” could be the “wide” filter. However, at the end the difference between wide and narrow isn’t …umm… exactly dramatic:
It looks like the passband is the same for both settings, just the skirts seem to drop a smidge faster in the narrow setting. The practical difference between the 2 settings is nothing to phone home about but certainly more noticeable in real life than the analyzer plot suggests, and also a frequent subject of complaint about later models of the PL-660. But the passband looks pretty good (flat), not too much of a drop-off until 2kHz, 3kHz is the the middle of the shoulder and the audio is ~12dB softer at this point.
The D-808 is a great example for the bunch of small DSP-radios that save extensive (and expensive) additional circuitry and offer a particularly rich set of (audio-) passbands for AM, even very unusual ones like “1 kHz”.
The passband is a little more sloped than that of the PL-660, but the tiny speaker likely keeps the impact of that in check. The big surprise is that the skirts are dropping so gently that even the PL-660’s ceramic filters seem to have a slightly better shape factor. I’m not sure if that is intentional or just a limitation of the Silicon Labs chips that these filters seem to mimic the shape of inexpensive ceramic filters, also in the next DSP radio in the group:
Given that this radio uses the same DSP chip in its 3rd IF stage, the difference in response is surprising: The slope within the passband is much more pronounced than that of the D-808 and 20 (audio-) decibels lower at the right/upper edge. The skirt is again surprisingly wide for a digital filter – both is also noticeable in the actual sound of that radio, often simultaneously a bit muffled and “too wide”.
Belka vs. Icom IC-705
Please note that the IC-705 uses the “oldschool” (IF-) bandwidth specification scheme, while the Belka is “per sideband”. Since both radios allow setting the lower/left skirt individually, I deliberately left the lower cutoff frequency at their lowest settings on both (I like to hear the bass in both voice and music!). Unlike the DSP portables, the right/upper shoulders are what I’d expect from an “ideal” digital filter.
The plot above also shows that the “4kHz” setting on the Belka is actually a bit wider than that, the Icom rolls off at 5kHz and the Belka at 4.5kHz (matching the European 9kHz MW channel spacing perfectly). In fact, all the bandwidths are tweaked 500Hz to 1kHz wider than the display suggests on the Belka, to fit the requirements of AM reception better. In SSB the displayed values are correct, it just doesn’t have separate display symbols for AM.
Now here’s why the Belka still shows more noise in its stopband: I made these plots at home, using an antenna to receive a signal (in lieu of a signal generator)…and interference: What you see there are the nasty idle pulses of a PLC modem in the neighborhood, a super wide signal covering the entire shortwave (minus the ham bands). The reason why you don’t see this on the Icom is that it has a pretty effective and adjustable IF noise blanker set to keep these pulses in check. All the other radios lack a noise blanker, and this visual demonstration shows why I wish they had one.
However, the additional presence of a true wideband signal was a good thing because I didn’t really think this through – the test signal should’ve been wider than the widest to-be-tested filter in first place. But all of the plots still tell their story, even when a small portion of the stopband noise seen on the other radios is caused by these pulses.
Extra-wide filters — nonessential but nice to have?
Besides the failed or semi-failed efforts to enhance and revitalize analog medium wave broadcasting – some AM stations exceed their allocated channel bandwidth (deliberately or not) and potentially have a little crispier audio, there are also experimental AM mode transmissions up to 32kHz wide on shortwave, all reasons why the demand for wide AM IF filters may have increased beyond the wish for best fidelity on standard medium wave.
The issue with wider-than-necessary filters (or filter settings) is that they do more harm than benefit on stations with a regular modulation bandwidth conforming to the standards. A slightly wider (or bad) filter can utilize the little extra bandwidth some stations claim for themselves and never sound dull, but they cannot produce treble that the station doesn’t transmit in first place, which seems to be a pretty common oversight. The surplus IF filter bandwidth will then only add noise/hiss and potentially adjacent channel interference, not fidelity.
However, the few actual wideband AM transmissions I have heard in my neck of the woods were quite impressive:
1. Radio China International is or was (to my knowledge) the only shortwave broadcaster doing experimental extra-wide (up to 32 kHz) transmissions on a somewhat regular basis.
2. Radio Romania Actualite is doing 16 kHz transmissions on medium wave 1152 kHz, basically occupying two Region 1-channels of 9kHz, leaving a little space left and right. They do that with 400kW (8 times the legal power of a Class-A station in the US) and the result is…niiiice:
This is an early evening signal from 820 miles away, almost sounding like an FM station next door! That almost makes me wish my favorite station would do that too! However, would these quite rare exceptions warrant buying a radio (meaning an SDR) for that?
Old man waves fist at… compromise filters
With the important parts of this post being said (you may skip the following oldtimer’s ramblings) – I couldn’t help remembering once again how good we have it today, well, at least in terms of IF filters:
It almost feels like yesterday that many classic general coverage receivers and most multiband portables had at best 2 (ceramic) filters with compromise bandwidths to do something useful in the shortwave 5 kHz channel spacing, in the MW 9 or 10 kHz channel spacing and on SSB. The “narrow” filter had to do for both SSB and narrow AM, so 3kHz was chosen on some radios to fit both of these purposes equally mediocre and – on the majority of all SWL radios – a 6kHz “wide” filter (meant to fit a clean 5kHz shortwave channel) had to suffice for medium wave as well.
Few of the more expensive tabletops also came with a 12kHz wide filter, the Kenwood R-1000 and Yaesu FRG-7700 are probably the best known examples, curiously their siblings R-2000/5000 or the FRG-8800 didn’t. As far as I can tell, none of the classic Icoms provided filters for what could be considered only really beneficial for local medium wave station reception anyway, but the JRC radios beginning with the NRD-505 did. This list may not be complete but it already covers most of them – medium wave audio fidelity or wideband digital shenanigans were not considered important enough by most manufacturers of SWL radios.
Since the provided filters were not always fully satisfying even for their intended main purpose either, aftermarket filter modification kits were quite popular back in the day, a bunch of dealers even sold readily modified versions of popular radios.
My favorite example of a radio that had to make due with only one filter is the Yaesu FRG-7, which was shipped with a ceramic filter specified with 6kHz (-6dB) bandwidth extending to 14kHz on the odd ‘-50dB’ stopband mark. In reality this specification turned out quite enthusiastic, that and the fairly big speaker gave the FRG-7 its reputation for rich audio …and awful selectivity. The same filter had to do on medium wave, shortwave and SSB, which then often covered 2 adjacent 3kHz SSB channels at once, with potential to have the chipmunks from 3kHz above/below completely drowning out the station you wanted to hear. That’s why many of the “frogs” were modified with one of the filter modification kits offered by several companies, the most popular probably being the kit offered by Gilfer in New Jersey.
Probably needless to say how amazing it seems in the light of the “we had nothing and had to push that uphill both ways” boomer lecture above, that even the most inexpensive DSP radios come with 5 different filter settings equivalent to 2, 4, 6, 8 and 12 kHz wide ceramic filters. My rudimentary analysis may confirm the perception that their digital replacements are not necessarily better, but none of the (digital) radios leaves you without a filter wide enough for best listening comfort on MW and SW, and several narrow ones to manage difficult reception conditions.
To keep the size of this post in check, I didn’t add a section to pick up newcomers and explain IF filter basics and details in depth. A good, short primer can be found e.g. here, a good explanation of terms like “shape factor” can be found here and if you want to dig into this more, here’s a blast from the past (1969).