What "Roofing Filter" means to Elecraft
Wayne Burdick, N6KR
There's been so much discussion about this topic that I'd thought I'd better try to clarify why we used the term when discussing the K3S.
A "Roofing filter" is simply a filter in the radio's first IF through which all signals must pass before they will be "seen" by later receiver stages. The narrower this filter is, the less exposure later stages will have. Thus a "narrow" roofing filter is desirable -- but "narrow" is relative, as I'll explain.
The term "roofing filter" has most often been used in relation to triple- or quadruple-conversion receivers. Such receivers have an IF above the highest RF band covered; it's typically something in the range of 30 to 70 MHz or higher. But "roofing" as a term should be interpreted as "protective," not "high in frequency." A roofing filter protects later stages, including amplifiers, mixers, narrower filters, and DSP subsystems, just as the roof on your house keeps rain out of all of the rooms. But a roofing filter can be equally at home at a low first IF, if that is how the radio is designed. It still provides the same protective function.
When we released the K2 in 1999, we never described our 1st IF crystal filters as roofing filters. We had only one IF, so the receiver model was simpler; there were no narrow filters at later stages that required protection.
But now, we find that the term is in widespread use. Average hams now think of roofing filter bandwidths as the standard of comparison between receivers. This is why manufacturers have jumped through hoops to try to provide the narrowest possible roofing filters. Many operators have an understanding (justified) that a roofing filter that is wider than the communications bandwidth will not best protect the receiver's later stages. So the term now seems appropriate to use even in a radio such as the K2, K3S, or Orion, all of which use low-frequency IFs (5 to 9 MHz).
In recent years, the roofing filter has become the centerpiece of receiver redesign:
Suppose that manufacturer "A" initially designed their receiver to use a 15- or 20-kHz roofing filter. Yes, this allows the receiver to handle NBFM and other wide modulation modes; it may also be selected to constrain the signal bandwidth ahead of a noise blanker or spectrum scope. But it comes at a price. If you're using CW mode, you'll have much narrower filters selected at the radio's 2nd and 3rd IFs. Yet the 1st IF roofing filter allows a broad swath of signals into the earlier stages. You don't need this energy in your passband. It can cause trouble.
Manufacturer "A," realizing they have a problem with dynamic range at close spacing, then announces that they've had a breakthrough: they can now offer a 6-kHz, or more recently 3-kHz roofing filter. This will certainly improve the situation for SSB and AM operation, but it still opens the barn door in CW or DATA modes, because the bandwidth is a factor of 10 wider than needed for communications.
So why don't they offer much narrower roofing filters that can be switched in for CW and data modes, or at times when adjacent-channel SSB QRM is very high? It's because they can't make filters any narrower at such a high IF.
Enter the "down-conversion" rig (K2, K3S, Orion, etc.). By converting to a low first IF, the designer can easily create narrow filters that are compatible with the required communications bandwidth. This is why we are offering filters with bandwidths as low as 200 Hz.
And yes, these are still "roofing" filters, because they limit exposure (bandwidth), thus protecting later stages (in the K3S case, the IF amp, 2nd mixer, and DSP).
73, Wayne, N6KR
K3S Crystal Filters and Dynamic Range
Eric Swartz, WA6HHQ
There has been a lot of discussion and speculation on the Elecraft email list about the necessity for narrow filters ahead of the K3S's DSP filtering and the impact of various filter bandwidths on receiver IMD. I'd like to clarify this topic a bit.
Our narrow crystal filters, especially the narrow CW filters, improve both 3rd order IMD and Blocking Dynamic range. IMD does NOT degrade when using the narrower CW filters. Using the DSP as the -only- narrow filtering stage will degrade IMD and Blocking dynamic range (inside of a wider crystal filter that preceedes it) for strong interfering signals above S9+25. We designed the K3S system to use both the narrow crystal and DSP filtering stages in series for optimal dynamic range performance. (Basic filter recommendations below.)
We've measured no serious degradation of IMD dynamic range when switching to narrower crystal filters on the K3S. We carefully designed the K3S to avoid this. We isolate the crystal filters from preceding amplifier stages to optimize return loss outside the filter passband. Going from the 6 kHz to 2.8 kHz to 400 Hz to 200 Hz crystal filters does -not- significantly increase IMD in the K3S. Just the opposite - Using our narrow filters improves both third order IMD and Blocking IMD at close spacing. As an example, we're seeing third order IMD DR numbers on 20M at 20 kHz spacing in excess of 104 dB and 97 dB at 5 kHz with the 400 Hz filter ahead of the DSP. Even better with the 200 Hz filter. These numbers are much worse if you do not use a narrow crystal filter ahead of the DSP.
We were very careful not to make the same mistakes made by other radio designs. Just as we did on the K2, which has a very similar first IF crystal filtering scheme, we have paid special attention to receiver gain stage balancing, proper isolation between the crystal filters and their surrounding amplifier stages and balancing the thresholds where hardware AGC and DSP AGC trade off their activation. Other areas of our design focus for optimal dynamic range are the PIN diode type and bias levels in the T/R switching area, the design of our front end bandpass filters, the core sizes used in those filters and inter-stage transformers and the IMD performance of the crystal filters.
In order to achieve the best K3S blocking dynamic range (desense) in the 140 dB+ range, you -must- use a narrow crystal filter (400 Hz, 250 Hz or 200 Hz for closer interfering signal spacing) in front of the DSP. We use hardware AGC after the narrow crystal filter and ahead of the DSP to protect the DSP when signals inside the crystal filter exceed a 100 dB dynamic range. If you only use the 2.7 kHz wider shape factor stock filter for CW or data operation you will be significantly desensed once signals within that filter's bandwidth exceed about S9+25. This is before phase noise from the transmitting station becomes a factor. Not uncommon on 40M at night, during a contest or at a multi-op station -- Or every day in major cities. As an example, changing to a 400 Hz or 200 Hz filter reduces blocking from signals 1-5 kHz away. I've personally confirmed this on the air with my K3S and the other commercial rigs we have here. When I've operated with the K3S, or another DSP rig, on CW without using a narrow filter ahead of the DSP filtering, I frequently experienced desense (BDR) from very strong nearby signals. Putting in the narrower crystal filter immediately cleaned it up. (While both the 400 Hz and 500 Hz filters are excellent CW filters, the 400 Hz filter performs slightly better than the 500 Hz filter due to its narrower shape factor. The 200 Hz filter is even sharper.)
Using narrow crystal filters ahead of the DSP also reduces AGC pumping from static crashes on 80/160M etc.
My personal real-world operating -basic- filter recommendations? In a nutshell:
SSB: 2.7 kHz or narrower shape factor 2.8 kHz
CW/DATA: 400Hz or 500Hz (plus 250 Hz or 200 Hz for some Data modes if you prefer)
AM: 6 kHz (and for wider SSB TX, we can select which filter you TX through and limit SSB b/w in the K3S's DSP)
FM: FM b/w filter (I believe its in the 12-15 kHz range.)
Add narrower/wider filters as you prefer. I like to use the 1.0 kHz crystal filter when tuning a crowded band or listening to a pile up. I use the 2.1 kHz SSB filter on RX when someone crowds me on SSB. We provide 5 crystal filter slots for the main Rx and sub RX to accommodate a wide range of personal operating preferences.
1. The stock 2.7 kHz filter is fine for most average SSB operation. Since we also transmit through this filter, for wider TX bandwidth and slightly sharper RX stop-band skirts you may prefer the sharper 2.8 kHz filter. The 2.8 kHz is also preferred for strong signal close in QRM and contest use.
For wider 'hi-fi' ESSB TX, you will need to transmit through the 6 kHz AM filter and let the DSP limit your ultimate bandwidth to something settable up to 4 kHz.
2. For most CW operation I recommend the 400 & 500 and 200 & 250 Hz crystal filters. I personally find the 400 and 500 Hz easiest to listen to for most casual CW operation, but I use the 200 Hz filter to dig out the weak ones when there is a lot of nearby interfering activity. For every day cw operation some prefer the 500 Hz to the 400 Hz due to the 500 Hz filter's slightly broader skirts. For the ultimate close in extreme signal blocking and IMD performance (2 kHz spacing and below) the 400 Hz and 200 Hz filters win out.
73, Eric, WA6HHQ