Oh, I can see lots of confusion forming here. Actually the issue of "roofing filters" is reasonably easy to understand. In fact, the Flex radios actually have roofing filters! (SURPRISE!) Now before everyone leaps on their keyboard to tell me how wrong I am I am going to add here that:
1. they are called something else; 2. they are put there for a different purpose. Let's go back to basics for just a moment and understand where and why a roofing filter is needed and why a roofing filter might actually be detrimental under some conditions. The purpose of a receiver is to receive a signal. (Basic enough?) Signals vary in strength. Your receiver can receive signals normally over a wide range of input power. At one end is the noise floor. Generally speaking you can't receive a signal whose power is less than the noise power in the same bandwidth. (No, let's not jump into a discussion of integration time and energy in coherent vs. non-coherent signals -- save that for another time.) At the other end is the point where the receiver starts to become non-linear, i.e. where the signal coming out is no longer an analog of the signal coming in. This is the compression point. At this point the receiver starts to generate a bunch of unwanted signals from the desired signals (intermodulation distortion or IMD). The range between these two points is the dynamic range of the receiver. (Yes, this is a simplistic description but please bear with me. We can get into the nuances some other time.) Now here is an important point: each stage in the receiver has its own dynamic range and its own contribution to nonlinearity and IMD. The dynamic range of the receiver is the combination of all the dynamic ranges of all the stages. We can improve things by making the receiver have fewer stages and/or making sure that the signal doesn't pass through a stage to contribute to the problem. (Keep this in mind, it will become an important point later on.) If there was only one signal to receive, we would be done but in HF and especially in ham radio, there are a lot of other signals besides the one we want to receive. Those signals are entering the receiver along with the desired signal. The combination of the desired and all the undesired signals passing through the receiver contribute to the RF power being amplified by the stages of the receiver. The trick is to remove as much of the undesired signals before they have a chance to saturate a stage causing compression and/or additional IMD. The first place we can do this is at the antenna. Resonant antennas do a good job of receiving RF power at desired frequencies. Signals at frequencies outside the operating frequency are attenuated so their effect on the total RF power reaching the receiver is reduced. The next place we can attenuate the undesired frequencies is the preselector preceding the RF-amp and/or first mixer. The goal is to eliminate, as much as possible, the undesired signals before they reach the next active (amplifying) stage. The only problem is, if the undesired signal is very close to the desired signal in frequency, neither the antenna nor the preselector will be able to fliter it out. The undesired signal will then have to pass through the following stages of the receiver. We have no choice. As someone else pointed out, the next place to get rid of undesired signals is in a filter immediately following the first mixer. If we have a single-conversion receiver, we call this the IF filter. It is the only narrow bandpass filter we have. It gets rid of everything except a narrow band of frequencies, those that we want to receive. If we have a multiple-conversion receiver (one with multiple IF frequencies) then this filter has come to be known as a "Roofing Filter". They call it that because the intention is not to use this filter to provide the final selectivity of the receiver but rather just to get rid of unwanted signals near the desired signal, that is, it shields or "roofs" the subsequent stages from most or all of the unwanted RF power. The final selectivity will be provided in a subsequent IF stage, usually at a lower frequency where selectivity is easier to come by. If you have a "traditional" receiver, i.e. one made within the last 20 years or so, it probably has at least two (double conversion) and maybe three (triple conversion) IF strips. These receivers often provide general coverage and have a first IF in the low VHF (up around 70MHz) in order to eliminate images. The problem with this high IF is that it is really difficult to make a very narrow filter. Getting a filter that is only 500 Hz wide at this frequency is very, very difficult. Most get by with a first IF "roofing filter" of about 20KHz. That means that if a strong undesired signal is within 20KHz of the desired signal, it makes it through the first IF and must be dealt with in the subsequent stages. This is why it was so difficult in the past to get receivers that had good close-in (less than 20KHz) specs, e.g. blocking dynamic range (BDR), intermodulation by the undesired signal (IMD DR3), etc. And given that, in a contest, the undesired signals may only be a few hundred Hz away, these filters do nothing to make things better. They may as well not be there. So to deal with this some people decided to go back to the original receiver approach that was used many years ago, i.e. use the lowest first IF that will get you adequate image rejection but where you can get really narrow filters. I believe that it was Ten-Tec and Elecraft that "pioneered" this retro-design. As a result, their receivers got rid of all that undesired RF power early on in the receive chain and they performed better. All they gave up was general-coverage. For most hams that didn't matter and it was a general win. You have only to look at the performance of these receivers to be convinced. But now most good receivers have DSP. DSP has some real advantages. It is possible to build ideal filters, amplifiers, attenuators, modulators, and demodulators mathematically. Frankly, they perform much better than anything that can be constructed out of analog components. The only problem is, given the state-of-the-art in analog-to-digital converters (ADCs) and processors, these functions must be performed at a relatively low frequency, i.e. tens of KHz or so if you want dynamic ranges in excess of 100dB. That means that we must convert again to a second, low IF so that the DSP can process the signal. The DSP then does the filtering functions and gives us the final filter shape with narrow bandwidths and very steep skirts. Anyway, it turns out that several of the the other "SDR" radios, e.g. the Elecraft K3, are really pretty standard dual-conversion superhet receivers, albeit very well designed. Yes, they have narrow roofing filters in the first IF but once your undesired signal is inside the passband of your roofing filter, it must get through that second mixer and the second IF amp to get to the ADC where it can be eliminated in the DSP. These additional components degrade the receiver performance. Their approach is to use narrower and narrower roofing filters to get rid of more and more undesired signal. If you want to spec your receiver at 2KHz spacing, make sure you have a roofing filter at 1.9KHz spacing to get rid of that signal 2KHz away before it reaches the 2nd mixer. This works, after a fashion, but eventually you reach a point of diminishing returns and that is -- analog filters aren't perfect. As you have probably experienced using very narrow analog filters on CW, you know that, as they get narrower they perform more poorly. They have increased loss, ringing, increased group delay at the edges of the passband, etc. Frankly, they muck up the signal. There is a point where they stop helping and hurt. In most cases you are better off using a wider roofing filter and letting the DSP be the sole provider of selectivity. This works because most of the time you don't need to get rid of a -20dBm undesired signal that is 200Hz away from a -110dBm desired signal. You can let the undesired signal(s) come through the first IF "roofing filter" to be dealt with by the DSP where you won't have the problem introduced by the narrow analog roofing filter. (This is why most people are throwing their money away when they fill their K3 or Orion II up with many expensive roofing filters.) Now we come down to the Flex radios and why they are different and why they don't really need roofing filters. First off, the only reason you need a non-zero first IF frequency is for image rejection. If you don't care about image rejection or if you can get image rejection some other way, you can use an IF of zero Hz. (Everyone has probably heard a good direct-conversion receiver and marveled at the clarity of the signal relative to a superhet.) Back before we had good, cheap crystal filters, people made SSB transmitters and receivers using the "phasing" method. This method used two separate RF and baseband (zero Hz) IF channels that are 90-degrees out of phase. By combining the signals from the two channels properly you could cancel out either the upper or lower sideband, the unwanted sideband being the image. These were the first I/Q radios (I = in-phase, Q= quadrature/90-degree phase). The difficult part of these radios that made them tricky to align and perform only moderately was coming up with the all-pass filters that introduced the 90-degree phase shift for the low IF. With DSP this is a piece of cake. So that is basically how the Flex radio works: RF from the antenna is split into two mixers (the Taloe or Quadrature Switching Detector) fed from the an LO that has two outputs in quadrature, converted to baseband, and fed into two ADCs which then comprise the digital I and Q channels for processing in the DSP. This gets rid of a LOT of extra hardware. If the hardware isn't there it cannot contribute to IMD. And the QSD, because it is basically just several on/off switches, is amazingly robust in the presence of very high-level signals. We don't need to worry about linearity because it isn't linear to begin with! So now we have only a few components between the RF signal and the DSP, i.e.: 1. the input bandpass filter; 2. the RF [pre]amp; 3. the QSD; 4. the first IF amp; 5. the anti-aliasing filter; 6. the ADC. This is almost the simplest RF/IF section you can build. There is very little there to contribute to reduced dynamic range and increased IMD because, well, there is very little there. Only receivers that dispense with the first mixer and connect the ADC directly to the bandpass filter are simpler but their need to do analog-to-digital conversion at RF frequencies limits their dynamic range as well. (Yes, I know we can narrow the bandpass filter, do subrate sampling, and then decimation to improve the dynamic range but I don't want to get into that here.) And I promised I would explain why I say that the Flex radios have a roofing filter. As you recall, I said that the roofing filter follows the first mixer to get rid of RF power that is outside the spectrum of interest. Turns out the Flex radios have this but they call it by a different name -- anti-aliasing filter. The anti-aliasing filter is there to reduce the power above the Nyquist frequency, i.e. half the sampling rate. Power above the Nyquist frequency is folded back into the passband. It is another sort of image. So the filter is there following the QSD. It is in the same location as a roofing filter and even performs many of the same functions. The only thing is, you can't change it. I suppose that an I/Q radio using a QSD could have anti-aliasing filters that track the sample rate or even cut off well below the Nyquist frequency in order to get rid of even more cruft before it reaches the ADCs ... just like a roofing filter. So, when you get right down to it, the receiver design in the Flex family of radios should be able to outperform other receivers when the undesired signal is inside the passband of the roofing filter. This is why Flex says that they don't care what spacing anyone uses between the desired and undesired signal for test purposes. The performance of the receiver is the same for all spacings. The other guys need to make sure that the "undesired" test signal is filtered out by the roofing filter in order to make their specs look as good as possible. And it is true that, if you can ensure that the undesired signal DOES fall outside the roofing filter, the other approach does work really well. After all, the K3, the Orion II, and the megabuck radios from Yaecomwood sport really good receivers. Just not as good as the Flex in my not-so-humble opinion. -- 73 de Brian, WB6RQN/J79BPL _______________________________________________ FlexRadio Systems Mailing List FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archives: http://www.mail-archive.com/flexradio%40flex-radio.biz/ Knowledge Base: http://kc.flex-radio.com/ Homepage: http://www.flex-radio.com/ Message delivered to arch...@mail-archive.com
