The subsystem that follows the frontend is the IF subsystem. In the down-conversion HF receiver presented here, the IF is situated at 9 MHz. One of the first decisions to make at this point is the architecture of the IF backend. My goal is to design equipment that although it can be connected to a PC, it still should be able to operate fully independent. So we are talking about a stand-alone box with all the controls on it! This rules out the economic and also very flexible option of a PC based SDR behind the roofing filters. In its most simple form this could have been a "SoftRock behind the Roofer" solution. This is no more than a Sampling Quadrature Detector (QSD) at 9 MHz with all the signal processing done by a connected PC in (Open Source) software.
For a stand-alone radio the options fall in the following broad categories:
In its simplest form this will be a single conversion superhet with analog IF followed by analog product detector and analog audio. This is the classic approach which still has many advantages. It is the architecture used for instance in CDG2000 and many other homebrew projects. In commercial radios this approach has been pushed away to the back ground when DSP became powerful and cheap enough to do complex signal processing more economically.
With the IF at 9 MHz there is little need for a double conversion superhet. Good quality quartz filters are available directly at 9 MHz. The complexity involved in a second conversion will probably outweigh the slightly better shape factors possible with for instance 455 KHz filters.
Analog + Digital
In practice this implies a double conversion superhet with 2 analog IF's where the second analog IF is at very low "ultra-sonic" frequency to allow easy digitization (ADC) and real-time signal processing in the digital domain with DSP. This is the approach followed today by most modern commercial radios and also some homebrew projects (Picastar for instance). The reason for the popularity of this approach is obvious: DSP is cheap, very flexible and very effective in mass production. Software running on the DSP is performing most core signal processing functions such as filtering, AGC and demodulation. Once the software is developed there is little or no cost involved in replicating it allowing greatly reduced cost in mass production. It basically comes down to the fact that silicon is cheaper than quartz and sometimes even better. The DSP software solution will also allow for more advanced signal processing functions such as for instance noise reduction and very narrow band phase linear filters which are hard or impossible to do effectively with analog hardware!
An alternative approach to the popular double conversion superhet could be a QSD behind the roofing filters to go directly to the baseband. The PC + Soundcard will have to be mimicked by an embedded DSP solution with quality ADC's.
Digital Down Converter (DDC) followed by DSP. This is actually the latest trend in Software Defined Radio (SDR), where the HF spectrum is directly digitized at the antenna with a very high speed ADC. The sampling rate, often between 100 MHz and 200 MHz, is way too high to be directly accepted by DSP for further processing. Therefore a step called 'decimation' is needed to convert down to a sampling rate that is within range for direct further processing with DSP. In SDR this is most often a PC connected through USB, FireWire or Ethernet. It does not take much imagination that this approach should also be very promising narrow band at 9 MHz behind the frontend roofing filters. This will take the heat away from the ADC completely and gives all the flexibility that comes with software to implement the rest of the signal path. Instead of a PC a standalone embedded DSP solution will have to be used to meet the goals.
It may come as a surprise that the choice of IF backend presented here at this point is not the fully digital modern DDC solution, but rather a conventional fully analog single heterodyne approach. First of all the economics of mass production do not play a role in typical homebrew "one-of-a-kind" developments. To design a fully digital solution including the software from scratch will take me probably more time than to develop a fully high performance analog solution from scratch. So with mass-production and money not as the key factors in the decision making, we have more options as homebrewers! DDC shows really great potential not only directly at the antenna, but certainly also behind the roofing filters of my frontend. However I feel that analog IF can still be very competitive and even further improved upon like turned out to be possible with the analog H-Mode mixer! I did design the analog IF backend presented here to be a baseline to compare fully digital backend developments against in the future.
Now that the difficult and also personal choice of what IF architecture to use is finally out if the way, what are the key factors that make up a high quality analog IF backend? Well, that can become quite an extensive list, but let us keep it top-level at this point and the following comes to mind:
Out of Band / Out of Channel Performance
The so hard earned dynamic range obtained with the frontend should be fully preserved, period!
In Channel Performance
The in-channel performance should allow for a very good sounding signal with as little additional distortion as possible.
The above two points do partition the IF subsystem into two separate modules (=boards) in the signal path in the following order:
The ultimate selectivity of the receiver has to be established before any significant gain is applied in the signal path. In case of a down conversion frontend, the required quality shape factor quartz filter is only just slightly narrower than the quartz roofing filters preceding it in the frontend. This makes the out of channel dynamic range almost as good as the out of band dynamic range. This is a serious advantage of down conversion radios, where this is at least technically possible. The up-conversion roofing filter at low VHF with similar intercept point and insertion loss is is almost impossible to manufacture with its much higher IF/BW ratio! As a result the VHF roofing filter will be much wider which causes an area inside the roofing filter and outside the channel that plays an important role in the close-in dynamic range of the receiver. The IP3 inside the roofing filter and outside the channel is usually much degraded with respect to the IP3 outside the roofing filter.
Good selectivity requires near rectangular shape factors or so called brick-wall filters. This comes inevitably with less phase linearity than with less rectangular filters. With CW this will be noticeable as ringing. With SSB it will sound more distorted. There is a compromise to be made between selectivity and good sounding audio! Phase linear filter design as much as selectivity allows is the key to good sounding signals.
Variable Gain Board
After the ultimate selectivity has been established the necessary gain can be applied to actually hear the signal in the baseband with the best possible signal to noise ratio and audio quality. This requires automatic gain control (AGC) with a wide gain control range as the frontend is able to cope with a wide signal dynamic range in excess of 120dB. When ignoring IMD, in-band SSB signals from MDS (-132dBm) up to at least 6dBm can be handled without difficulty by the frontend. So this sets the tone for the gain control of the IF as well.
In case of SSB and CW, actual signal strength or envelope is obviously very dynamic by its very nature and this requires a well designed AGC control loop. This is the dynamic behavior of in channel signals and the AGC loop dynamics will determine how well the signal will sound at the speakers. AGC action does produce IMD by definition which can seriously worsen the so hard earned good in-channel IMD performance of the preceding stages.
How to characterize IMD
The pages related to Antenna Band Pass Filters and Frontend Board are using the third order intercept point or IP3 extensively to characterize the IMD3 performance of these subsystems. IP3 is usually expressed in dBm and is a theoretical figure that describes the level where two equally strong input tones will produce equally strong IMD products in a non-linear system. The whole idea of using IP3 for this depends on how well the system under test behaves with respect to IMD. If 3rd order law applies to the DUT, then the calculated IP3 figure will be independent of signal level. That makes the IP3 quite a useful figure for easily comparing performance and also other figures can be easily derived from it:
The third order dynamic range or IMD3DR can be calculated from IP3 and MDS with the following formula:
IMD3DR (dB) = 2 * ( IIP3 - MDS ) / 3
For example, a receiver with an IP3 = +50dBm and MDS = -130dBm will have IMD3DR = ( 50 + 130 ) * 2 / 3 = 120dB.
The absolute level of the IMD products can be calculated given the IP3 and the level of the input tones with the following formula:
IMD (dBm) = 3 * Pin - 2 * IIP3
A receiver with IP3 = +50dBm and the 2 input tones have both Pin = -10dBm, the IMD products will be 3 * -10 - 2 * 50 = -130dBm strong. This is right at the MDS for the receiver in the previous example. The IMD at MDS level is -120dBc, which is exactly the receivers dynamic range.
From the above result it can be easily seen that the relative level of the IMD products with respect to the input tones is given by the following formula:
IMD (dBc) = 2 * ( Pin - IIP3 )
A receiver with IP3 = +50dBm and the 2 input tones have both Pin = -10dBm, the IMD products will be 2 * ( -10 - 50 ) = -120dBc strong.
This is all very convenient, however if the DUT does not follow 3rd order law perfectly then the IP3 figure becomes far less meaningful as its value also depends on the actual input level where it is extrapolated from. A complete table of IP3 values with corresponding input levels would be needed to fully characterize the IMD performance of the DUT. In that case we might as well forget about IP3 and just list the absolute IMD levels against input levels, or relative IMD below carrier against input levels.
The IMD generated inside the IF subsystem is actually only in-band IMD thanks to the protection given by the roofing filters on the frontend board. As the roofing filters are about as wide as the signals involved (2400Hz SSB, 500Hz CW), the IMD introduced by the IF subsystem before, inside and behind the quality filters is really only in-channel IMD. In-channel IMD is IMD that by definition co-exists with the in-channel signal that causes the IMD in the first place. This is quite a different situation from the out-of-band IMD on the frontend board, where the IMD can be heard inside the channel, but the signals that cause it are outside the channel! For in-channel IMD, IP3 is not a very useful way of characterization. Far more meaningful are the dBc values against the input values. We know that if the IMD is -60dBc, it will not be very significant with respect to the desired signal. In audio terminology this would probably qualify as HiFi. A value as bad as -20dBc however will be noticeable! Note that 100W - 150W solid state LPA's used by amateurs are producing IMD around -30dBc and sometimes worse. The better tube amplifiers may produce as little as -40dBc IMD.
In-channel IMD in this discussion is in the end just distortion of the signal at the speakers of the receiver. It will not play a role in the dynamic range of the receiver. For the IF subsystem this discussion is also quite relevant because the in-band IMD for quality shape factor quartz filters is usually far from 3rd order law perfect. Also the IMD produced by the variable gain AGC'ed stages, where the output level is kept constant, does not follow 3rd order law at all. So now that both cases apply to the IF subsystem, non-3rd order law behavior and in-channel IMD, the IMD in the IF subsystem is best described with dBc values.
Our design goal will be to have the IMD introduced by the IF subsystem to be at a level of -60dBc or better.
Please follow the links below to read more about the Selectivity Board and Variable Gain Board that together represent the 9MHz analog IF.
Variable Gain Board
Back to Receiver Overview
Back to the TOC