Online Postgraduate Courses for the Electronics Industry

Provided By The University of Bolton

• Home
• About Us
• Questions & Answers
• Courses
• Students
• Links

Useful Tools

Communications IC Architectures

Unit 1: Introduction to Communications Circuits and Radio Architectures

Unit Contents

• 1.1 Introduction
• 1.2 Basic Requirements for an HF Radio Receiver
  • 1.2.1 Dynamic Range
  • 1.2.2 Sampling Requirements
  • 1.2.3 Filtering
• 1.3 ADC Constraints
• 1.4 Digital Signal Processing Issues
• 1.5 Architectures
• 1.5.1 Zero I.F.
• 1.5.2 Single I.F
• 1.5.3 I.F. Sampling
• 1.5.4 Fast Fourier Transform (FFT)
• 1.6 Receiver Design Example
• 1.6.1 Front End Filtering, Mixing and LO
• 1.6.2 IFs and the Analogue Stages
• 1.6.3 Analogue to Digital Conversion
• 1.6.4 Digital Filtering
• 1.6.5 Digital Demodulation
• 1.6.6 Audio and AGC
• 1.6.7 Control Functions
• 1.7 Conclusions
• 1.8 Appendix 1
• 1.9 Appendix 2
• 2.0 References

Preface

Preface This chapter assumes a basic knowledge of radio systems design. A good text for background reading is the Radio Communications Handbook, by the RSGB, now in a new edition. This is strongly recommended where the student feels that he or she needs more background than is provided here. Click here for a Glossary of Terms.

1.1 Introduction

Although most professional radio receivers built in recent years could be claimed to be digital in the sense of having digital synthesisers and sometimes digital control functions, digital receiver technology is usually interpreted as the provision of at least some of the selectivity functions of the radio in digital form. By this definition, the first "digital" radio to appear commercially was the Rockwell-Collins HF 2050, which was reported in reference 1. Since the early 1980s, there have been a number of papers leading to the digital realisation of radio receiver functions, both at HF (refs. 1 to 5 and 8 to 13) with others aimed at VHF/UHF (parts of ref. 3, 6 and 7). Modes are primarily SSB (refs 10-13) or FM (refs 6 and 7) although most papers concerned with HF include recognition of other modes such as FSK, PSK, DSB and ISB. CW is still included since it is still much used by the merchant navy of Eastern European countries. Note here that digital techniques may be used for analogue radios, ie. where the signal is transmitted purely in analogue terms, and equally analogue techniques are used in some stages of radios where the transmitted signal is digital.

The motivation for taking the digital approach varies. It may be cost, size, performance or flexibility. However, it is important to note that there have been significant advances in analogue design in the last decade, especially in the context of high performance HF radio, and it is essential to recognise that performance of the very best analogue designs may not be improved upon by digital techniques. What digital techniques can offer is a combination of advantages in all these areas, with a reproducibility which cannot be approached by all analogue systems. There are also some areas of receiver design which can be carried out uniquely well in digital form, especially in the pass band phase response of filters; but it is in the stop bands of these filters where the analogue equivalents are better, so the advantages are less clear.

Simplistically, the "ideal" digital radio receiver would consist of an analogue to digital converter connected to the antenna, with a following DSP device providing all the signal selection required. There may be a few applications where this will soon be practical, eg. microwave surveillance receivers, where the key parameter is very fast response, and correspondingly wide bandwidth. However, for the HF bands, the large number and wide amplitude range of received signals is beyond the capabilities of present generation ADCs. Reference 15 and table 1 (in the Appendix) indicate the levels of signal to be expected, even in a relatively quiet environment, on the HF bands. Urban environments currently show very high levels of HF and VHF signals due to the presence of many accidental or deliberate RF emitters. This situation is unlikely to get better, despite being the subject of much legislation, and will probably get worse, especially as new services come into existence in the VHF, UHF and microwave bands. Most notably of course is the frequency range around 900MHz, which 10 years ago was a quiet part of the spectrum and is now heavily used by mobile phones. The same is also true of 1800MHz. The design constraints below apply in principle to all frequencies. There remains the subjective difference that UHF and microwave signals do not propagate beyond line-of-sight except in the most unusual circumstances, while VHF signals may operate just beyond line-of sight. On the HF bands, however, world-wide communications are possible without repeaters. In contrast, HF transmitters of high power are relatively rare, while UHF/microwave transmitters are now so common that all circuits must be designed on the assumption that a transmitter is relatively close at hand.

[back to top]

1.2 Basic Requirements for an HF Radio Receiver

1.2.1 Dynamic Range

The basic problems are the dynamic range required and the frequency response. The maximum theoretical signal to quantisation noise ratio which can be obtained from an N bit ADCis:

S/N = 1.76+6.02N dB or, for a 16 bit converter, 98.08dB.

In a practical receiver, the smallest usable signal must be say 10dB above the quantisation noise to be recognisable, so even this 16 bit ADC would only have a real dynamic range equivalent to 88dB from recognisable signal to serious overload. Again, in the practical case, it might be felt desirable to avoid any possibility of ADC overload since, unlike the corresponding analogue circuits, ADCs go into severe non-linearity at the point of overload.

1.2.2 Sampling Requirements

The Nyquist sampling theorem indicates that a signal must be sampled at more than twice its highest frequency component to avoid ambiguity. This can be seen in the simple example of figure 1.

Figure 1 Nyquist Sampling in the Time Domain

Figure 1 shows sampling in the conventional way, i.e. at the limit of 2 samples per Hertz of input. In most cases, although this is the theoretical limit, it is impractical, since the phase relationship between the sampling pulse train and the input signal is undefined. This can lead to the sampling occurring at exactly zero on the input waveform or at exactly the peaks.

For any sampled waveform, there is a sampling loss of amplitude which can be quantified as (sin x)/x, where x is the relative frequency.

For a practical receiver, the spectrum of signals is sampled at more than twice the highest frequency signal, as shown in figure 2.

Figure 2. Nyquist Sampling in the Frequency Domain

Figure 2 shows a range of input signals from zero to a frequency just below half of the sampling frequency Fs. After sampling by Fs, the output will contain frequencies as at the input, and their images generated by Fs. There will also be higher 'alias' frequencies as noted below.

The sampling process generates an image frequency response, indeed, many of them as shown in figure 3. Sampling can be expressed mathematically as the convolution (multiplication) of the two signals. In this case, a signal is sampled at much less than the centre frequency. However, the Nyquist theorem refers to the information bandwidth in the signal, so sub-sampling in this manner is permissible without ambiguity. The bandwidth of the signal, including both wanted and any present but unwanted components, must be limited in such a way that the twice-maximum frequency criterion is always met. If it is not, then aliasing of components into the frequency space of wanted components is inevitable and cannot be removed later.

Figure 3, Convolution of an IF waveform

1.2.3 Filtering

The filter bandwidth required for each of the radio modes of operation is critical to obtaining the best signal to noise ratio, and signal to interfering signal level across the bands. The usually accepted bandwidths are shown in table 2 in Appendix 1. Filters as wide as 12kHz may be used to include some forms of FSK or older standard FM. In many modern radios, a first filter (the "roofing" filter) provides initial selectivity at say 12kHz wide, and with very good skirt response.
 

Figure 4. Filter Terminology

Figure 4 illustrates the filter terminology. Important characteristics of a filter are the bandwidth, the insertion loss, and, especially important on multiple access bands, the sidelobe suppression level. Very often, crystal filters, which still dominate this market, have low or non-existent sidelobes. Cheaper ceramic filters do have some sidelobes, while in the case of digital filters sidelobes are likely to be the main problem. Clearly, a high rejection of close in signals is desirable, but not if the performance further out in frequency is compromised.

The combined effect of analogue and digital filters in a radio is illustrated in figure 5. Here it is assumed that the digital filter has alias pass bands not far away from the required pass band, as in figure 3, so an analogue filter is used, as in table 2, to provide an overall roofing function, usually at I.F. The alias pass bands are therefore rendered unimportant in the response. Sidelobes on the digital filter may still be important however, since they could occur within the roofing filter pass band; this is the reason for many of the design compromises required, and will be extensively described below.

One further effect of the roofing filter is to provide a high ultimate rejection, again an area that is difficult in digital form. Unfortunately, the aggregate response owes so much to the analogue components, in particular the filter, that it can be argued that the digital filtering is superfluous in a really high performance receiver, if compared to state of the art crystal (analogue) filters. This is strictly true at the present, but ignores the actual advantages in cost, size and reproducibility of the digital filter. In addition, digital demodulators can be far more versatile than their analogue counterparts. Where the transmitted and receive signal is itself digital, the case for early digitisation in the receiver is made.

Figure 5 Effect of Analogue and Digital Filters

[back to top]

1.3 ADC Constraints

Several approaches to this problem have been considered. The first (ref 1) uses a 7 bit ADC sampling a 3MHz I.F. at 12MHz. This data is then re-sampled digitally to form two 16 bit 48kbps channels (i.e. I and Q). These signals are digitally filtered and decimated into two 16kbps, 16 bit channels before digital demodulation. A particular problem with this approach arises from the ADC accuracy limitations, both integral and differential. Typically, ADCs specified for wide bandwidths offer only small word widths, such as 7 bits in this example, 8 bits elsewhere, and rarely more than 10 bits for a few tens of MHz sample rate. The reasons for this are inherent in the structure of fast ADCs, and while newer structures are constantly being reported, they usually have advantages in speed and power consumption rather than word width.
 

Most ADCs are specified to have an accuracy, both differentially and integrally, which is similar to, or only slightly better than, their resolution. Thus, an 8-bit converter might have integral and differential errors at the 9-bit level. We can consider the two types of error as approximately separate issues. Integral linearity errors will lead, in the presence of full scale signals, say from an unwanted source, to output components at the frequency of the source and harmonics of it. Of course, if there are two or more such sources, the output will contain all the harmonics and mixing products from all of them. This could be a serious problem but one that can be alleviated by input, necessarily analogue, bandpass filtering.

Differential non-linearity shows a related problem. Suppose that the wanted signal takes up just a few levels of the ADC, as it must in the minimum case. Then if these levels include a non-linearity, even of just one level, it represents a serious error in quantisation of the very small signal. If there is also present an unwanted signal, even if of smaller magnitude to the wanted one, it will, due to the non-linearity of the response, modulate onto the wanted signal. In the general case, this cannot be removed by subsequent filtering. The only means of avoiding this problem is to provide all the required filtering ahead of the ADC, in analogue form. The digital functions are then used primarily as a multi-mode demodulator. This need not be such a great disadvantage as it seems; the provision of a relatively narrow analogue "roofing filter" of this type will be seen to be very desirable in most classes of digital radio receiver.

To better use the digital capabilities therefore will require an ADC of 16 bits or more; some converters of 14 bits have been used, but must be seen as a compromise in design. The ADC, which will therefore be limited to a few 100KHz sample rate or less, must be preceded by some form of down conversion; this dictates the form of the front end of the radio architecture.
 

Sample and Hold

A related function to the ADC is the sample and hold; sometimes this function is described as a track and hold. The functional difference between the two is small, and really rests on whether the device has sufficient time to actively track the input signal or whether it acquires the input value and then holds it. Commercial devices are not consistently described in this aspect.

The sample/hold function is necessarily analogue in form. Under the control of a digital clock, the input signal is sampled by a switch device, and that value is held for a predetermined period, during which the held value is processed, typically by analogue to digital conversion. A properly designed sample and hold can sample very accurately and, especially importantly, at a very closely controlled time. The sample clock rate can be set to a very high specification to avoid sample to sample jitter in timing; figures as low as 1ps have been claimed. It is then possible to sample an input frequency, say the I.F. of a radio at 10.7MHz, but only to use samples at a fairly low rate, e.g. 100kHz, in the ADC.

The sample/hold therefore performs the required down conversion prior to the ADC. A refinement of this technique is on-frequency sampling. In the radio architectures published, where sampling is used, it is at a fixed frequency, and all inputs are treated equally. In practice, the one thing that is known about the signal to be observed is the frequency. A better architecture is therefore to sample at the wanted frequency, rather as in the direct conversion receiver approach. The difference is that, with sampling, it is possible to average many samples from the S/H before going into the ADC. A circuit for averaging and discarding old samples can be arranged, if necessary by multiple sampling. Then, this circuit can be seen as a first stage filter in the process. Because many samples are present, an analogue equivalent of an FIR filter can be arranged, providing good phase characteristics and good roll-off ahead of the main digital filtering.

As described, this filter is really only suited to SSB, where phase coherence is not a problem. A further refinement would be to use sampling at four times the wanted frequency. Although demanding on S/H performance, this would give very simple I/Q operation, while the averaging facility described above would be further enhanced. Of course, the sample/hold device must have acceptable characteristics is terms of sampling speed, accuracy and factors such as "droop", i.e. loss of signal level during hold. High frequency S/Hs also have parameters such as isolation, feedthrough etc., all of which tend to limit accuracy, and which are often difficult to consistently determine from data sheets. Practically, sample/holds of 12 bits equivalent accuracy are available to 10MHz sample rate.

The alternative to sample/hold down conversion is the conventional mixer type down converter. This has the virtue of simplicity, although this is offset to some extent in I/Q systems. Essential features for the mixers are as in any high performance receiver, i.e. very wide dynamic range with good port to port isolation. Such mixers are currently fairly readily available.

[back to top]

1.4 Digital Signal Processing Issues

Digital filtering and demodulation are very large topics in themselves, and can only be touched-on here. However, certain generalisations can be made, and some conclusions drawn.

Clearly, since dynamic range of the radio is a major issue, the DSP system chosen must not be more restrictive in this respect than the ADC. This should not at first examination be a problem, since there are many DSP chips available (see tables 3, 4 and 5 in appendix 2) which offer 16 bit word length, or, in a few cases, more. Clearly, more would be desirable, and it is essential to include in the DSP sufficient provision for word length growth in computation, either by provision of very long words internal to the DSP, or provision of floating point.

Broadly, there are two main classes of DSP chip, respectively the microprocessor-like devices (table 3) which offer custom control and programming via a conventional micro, and the dedicated custom DSP parts for (usually) FIR filtering (table 4). The key difference is that the former group are very flexible in their programmability, but contain usually only a single hardware multiplier/accumulator, and require multiple passes to achieve filtering, while the latter group consist of several (or many) MACs. Recently, a "digital down converter" device has been introduced, specifically for radio applications. (table 5).

Figure 6 Four Tap FIR Filter

The programmable devices can, ultimately, offer unlimited filter complexity, i.e. many "taps" in the FIR filter sense. They do this at the cost of speed. If a single MAC takes 50ns, then an 8 tap filter, say very roughly equal to a 4 pole analogue filter, needs at least 400ns. Additional delays in the circuit would probably extend this to 500ns, or 2MHz maximum throughput rate. A two channel (I and Q) system would of course halve this rate. Correspondingly, a dedicated high speed DSP might offer 32 taps and >10MHz sample rate (see table 4) at the cost of limited flexibility and some complexity of programming.

A simple FIR filter is shown in figure 6. The data paths are all "N" bits wide, where N is at least 16. For the delay elements, this is not a problem, since it needs only 16 latches. However, for the multipliers, 16 x 16 bit multipliers are not trivial. It is usually possible to discard the lower 16 bits of the output, having made provision for word growth, but these elements represent a significant amount of hardware. In the DSPs of table 3, this is resolved by using a single multiplier and many passes, under software control. In the DSPs of table 4, several multipliers are incorporated.

There is one further solution, which is just becoming available. This is a single chip digital down converter (DDC), which contains all the functions required for phase splitting, first stage digital filtering further down conversion and two (I and Q) 121 tap FIR filters for low pass filtering. It will be obvious from the above comments that this filter cannot, on a practical chip size, contain fully specified multiplier coefficients, since this would require two hundred and forty two ( for two channels) 17 x 17 bit multipliers. The more likely explanations are either that unity coefficients are used, or that the coefficients are binary, so that simple shift and add functions are possible. A dynamic range of 102dB is claimed for the part, which can be reconciled to the 16 bit data input word, because a two channel approach is used, adding a further 6.02dB.

As a digital part, no error specification is included; in a practical system, the preceding ADC would (rightly) be the operational limitation. Selectivity bandwidths of 659kHz to 322Hz are possible, in architectures similar to those shown below. Infinite impulse response (IIR) filters are also a possibility for radio systems; they differ from FIRs in having feedback paths. This usually means that they are less complex for a given selectivity, but the feedback paths both reduce the bandwidth available and introduce the possibility of oscillations, or ringing. IIRs tend not to be used in current generation digital radios, but this may change.

[back to top]

1.5 Architectures

1.5.1 Dynamic Range

The simplest digital receiver architecture is the "zero I.F." or direct conversion approach shown in figure 7. To achieve reasonable bandwidth, the incoming signal is heterodyned down to zero frequency (sometimes referred to as "D.C.", or "baseband") using an analogue mixer and local oscillator at the same frequency as the input. The ADCs then operate on this signal, typically in the phase and quadrature system (I and Q) shown, in order to obtain as much of the input information as possible. The DSP then provides the filtering required, resolves the sideband ambiguities and converts the signal into a usable form.

This latter conversion corresponds to detection in a conventional radio. All modes of signal can be demodulated in this way, although FM presents particular problems, since the receiver must operate linearly right down to literally zero frequency. The I-Q system offers resolution of phase ambiguities so it is convenient to visualise a pass band in the frequency domain from positive to negative frequencies. The filter characteristics required will therefore be half of the audio bandwidth required, say, for a 3kHz audio bandwidth, the low pass filters would be required to roll off at 1.5kHz.

Capacitor coupling between stages after mixing to baseband (there may be several gain stages before the ADC) is not permissible, since it would distort or remove the zero frequency response, leading to a "hole in the middle" of the pass band. This would be disastrous for analogue FM reception. The saving feature of this structure is that FM is rarely used on HF, so is not likely to be a problem. FSK can be acceptable, even with an A.C. coupled direct conversion receiver, since only the extreme states are used; it is usually possible to avoid the hole in the response.

The most important HF mode for communications is SSB, and this is the most easily demodulated mode for this class of receiver; even the phase ambiguity removal is not strictly necessary, although it is desirable in order to "phase out" and signal in the unwanted sideband. It would also make sideband selection easier, and facilitate the detection of DSB and SC-AM. The fundamental problem with this class of receiver is that the sidelobes available with practical digital filters are unlikely to be acceptable for a high performance radio system. Typically, a spurious free dynamic range of at best 70 dB would be expected.
 

Figure 7 Zero I.F. Receiver

[back to top]

1.5.2 Single I.F.

A further development of this class of receiver is shown in figure 8. In this case, an IF roofing filter, as mentioned above, sets the overall frequency response, and in particular determines the ultimate rejection of the filter system. The second mixers convert the signal down to baseband, where it is digitised in the same way as the zero IF system.

The major difference between the two systems is that the selectivity is mainly provided by the analogue, typically crystal, filter in the single IF system. This will give a spurious free dynamic range of up to 100dB, for signals outside the pass band. Generally, this filter will be 5kHz-50kHz wide. Clearly, any signals within this bandwidth rely on the digital filter for selectivity. Again, sidelobes are an issue, so that the design becomes a compromise of flexibility from the digital filtering against dynamic range from the analogue filtering.

Figure 8 Single I.F. System

1.5.3 I.F. Sampling

Figure 9 shows another variation on the theme of IF digitisation. In this example, the IF is sampled directly, i.e. it is not converted to baseband first. This has the advantage that only a single ADC is needed; the I-Q split can be carried out digitally, and to much greater accuracy than in analogue form. The IF will necessarily be fairly low, to get within the range of available ADCs of high accuracy. Low accuracy/high frequency ADCs introduce a serious compromise of dynamic range, although if the roofing filter is narrow enough (ref. 1) this may be acceptable.

Figure 9 I.F. Digitisation

1.5.4 Fast Fourier Transform (FFT)

In principle, it would be possible to take the output from a fast ADC and Fourier analyse it using a fast Fourier transformer. This technique has been proposed for the VHF and UHF bands, either with the ADC working directly at the input frequency, or at an IF. The ADC performance is a compromise against usable dynamic range. For a high speed ADC the limitation is, as explained above, probably 8 to 10 bits. This would give only 48dB-60dB dynamic range, which may be suitable for a microwave radar FFT, but is unlikely to be adequate even in the VHF and UHF bands.

On HF, as indicated above, dynamic ranges of 100dB or more are very desirable. One approach to this is to take a 16 bit ADC output, sampled at say 100kHz, and perform a low speed FFT on this. This would give a spectral coverage of ±25kHz on the nominal frequency. The input frequency can then be rapidly stepped to give wider coverage, for instance using a direct digital synthesiser, probably within a phase locked loop.

The upper limitation in sweep time is easily calculated for any receiver. The final bandwidth of the receiver is determined primarily by the mode of operation. Then, in order to obtain any information at all within the channel, at least one Hz of the signal in that channel must occur as the receiver sweeps through the frequency. Thus, if the bandwidth is 3kHz, the sweep rate must not exceed 3kHz/sec. This will of course recover only the information that there is a signal present. Slower rates will be needed to determine actual signal characteristics. Sweep rates of this type are well within the capabilities of modern synthesisers.

Where the signals are fading or intermittent, longer observation periods are necessary.

1.6 Fast Fourier Transform (FFT)

A Receiver Design Example Figure 10 shows the architecture of a practical receiver. This is based on reference 12, which describes such a receiver, although the comments here extend this work further. Although this design is based on HF requirements, in many respects these are carried through to VHF, UHF and microwave systems. The caveat is that such systems, to be hand portable, must be low power and especially low cost.

Figure 10 Receiver Example

1.6.1 Fast Fourier Transform (FFT)

Front end Filtering, Mixing and LO The first stage of the receiver is a sub-octave analogue filter bank. While not fundamentally essential, front end filtering very greatly reduces the probability of large interfering signals in band; see table 1 and ref. 15.

The first mixer, typically a Schottky diode ring mixer, should provide very wide spurious free dynamic range. To achieve a reasonably low noise figure, the mixer may be followed by a high signal handling gain stage, or the mixer may be a fully active type. No R.F. amplification is shown; although this may be desirable at the top end of the HF bands, say above 20MHz and is usually mandatory for VHF and UHF. The incoming noise levels below 20MHz mean that RF amplification is unnecessary and would reduce large signal handling.

A single reference source locks both frequency synthesisers together for best phase performance. The synthesiser can be a straight phase locked loop (PLL), a multi-stage PLL or a combined PLL and DDS in order to gain fine frequency incremental control. Alternatively, small frequency increments are available in some forms of DSP.

1.6.2 IFs and the analogue stages

The choice of IF frequency is determined by a number of factors. There are standard frequencies such as 455kHz, 9 MHz, 10.7 MHz, 21.4 MHz, 45 MHz, 70MHz and 90MHz. The advantage of a standard frequency is in availability of filters at reasonable cost. Of these, clearly 455kHz is impractical as a first IF, for image rejection reasons, although it may be suitable as a second IF in a triple conversion system; this is the method used in ref. 1.

Nine, 10.7 and 21.4 MHz would all be suitable in respect of image rejection, but would necessitate avoidance of those frequencies. For certain classes of receiver, including some military receivers, aeronautical receivers or others where only limited band coverage is needed, any of these would be a good choice. For a general coverage HF receiver, however, it is now conventional to up convert to an IF of 40 - 90MHz.

There are some advantages in going to the higher end of this range, i.e. avoidance of harmonics of in-band signals, but this consideration is marginal. A choice of 45MHz would suit most purposes. An advantage of the up conversion is that the local oscillator need cover less than one octave, i.e.47MHz-77MHz for 2-30MHz coverage and a 45MHz IF. The IF bandwidth of this filter should be as wide as the widest mode in use, say 8kHz. It would, very desirably, be much wider, say 50kHz, but this involves the compromise mentioned above of the relatively poor sidelobe rejection of the digital filters. The new chip referred to in table 5 may at least partially overcome this problem

After the IF filtering and amplification, the signal is I-Q split at the second conversion stage. Appropriate phasing of the response, and channel matching, can give 40dB suppression of the unwanted sideband; further suppression is available from the DSP.

1.6.3 Analogue to digital conversion

The ADCs, as pointed out above, will give 98dB range, but this should be reduced to 88dB for recognisable speech. This is the limitation on linearity within the filter pass band; outside the pass band, the analogue circuits determine the linearity of the system.

The noise floor in a 2.4kHz bandwidth, typical for SSB, is kTB=-141dBm. In the lower HF bands, the atmospheric noise is typically 30dB to 40dB worse than this; around 7MHz in the evening is many times worse again due to the number of signals present. One design approach is to set an arbitrary noise figure for the receiver of say 10dB, which is adequate for the upper HF band, above 25MHz. Then the noise floor is -131dBm, and the minimum signal as above is 10dB above the noise floor at -121dBm. This will be more than sensitive enough at the lower part of the band; it may be desirable to attenuate the signals lower down in frequency, or to use a less efficient antenna.

The local oscillator must also be of adequately low noise, typically -121dB/Hz at 5kHz offset for this class of receiver. This point is often missed in receiver design, and leads to the phenomenon of "reciprocal mixing", (see figure 5) i.e. signals mixed with noise come into the IF pass band. The local oscillator therefore needs to be very clean in this respect. Frequency synthesis is a major topic which is addressed in other modules of this course; it is critical to good overall receiver performance.

1.6.4 Digital Filtering

The DSP devices set the main selectivity of the receiver. Bandwidths can be set from say 300Hz for CW reception to 6kHz or more for AM. SSB is usually set at 2.4kHz, although the facility for reducing this under difficult circumstances can be useful; intelligibility suffers below 1.8kHz, even for the experienced operator.
 

Receiver gain should be minimised, but some gain will be needed ahead of the ADC. This must be as linear as possible.

The DSP device sets the ultimate bandwidth of the system, subject to the limitations of reciprocal mixing as above. One problem of digital filters is that truncation of the digital filtering process, i.e. the finite number of taps in the filter, gives rise to many sidelobes in the frequency response. There are many different "windowing" functions available for the calculation of the filter coefficients which minimise this effect, but it is to be expected that finite truncation effects will produce sidelobes which are not better than 90dB below the signal; -70dB is regarded as good. This is clearly a problem. The most satisfactory approach is to use a very long FIR filter; this is now becoming possible with devices such as the HSP50016 (table 5) which offers two 121 tap filters and 102dB dynamic range.
 

1.6.5 Digital Demodulation

Also in the system DSP, although probably in a separate chip to the filtering, the demodulation can be arranged to operate in any mode. SSB is relatively simple, since the conversion to baseband effectively does the function. Fine tuning can be carried out in the DSP if necessary. Phasing out of the opposite sideband is desirable, although not essential if the filtering is adequate.

AM is only a little more difficult. Removal of the carrier needs only a D.C. block, then the signal is treated as SSB/ISB, with phasing required for the latter mode, when both sidebands are present and different.

FSK and PSK are demodulated by comparison with reference digital values. True FM, where all frequencies can be present, is demanding of this type of discriminator, since the response must remain level down to D.C., which can be a challenge in the balancing of the two halves of the circuit. An alternative is to offset the FM centre frequency and to demodulate it as a counted output away from a reference. This can be arranged to operate in an A.C. coupled manner. CW and some data modes can be demodulated as for SSB, and can benefit from the narrower bandwidth. Software is needed for control of frequency selection, filter bandwidth and filter characteristics. The availability of software routines may be a major factor in the choice of the DSP part.

Automatic mode selection has been demonstrated using neural networks, and will become an option on radios where rapid identification of mode is desirable (Ref. 13).
 

1.6.6 Audio and AGC

Audio reconstruction is easily achieved using a non-critical DAC. Automatic gain control (AGC) is essential in any high performance receiver, and in this case must be a mix of analogue and digital methods. The key issue is to derive the AGC control function from the wanted signal, rather than from some combination of wanted and unwanted.

The preferred source is therefore the demodulated, but still digital, output from the DSP. The priorities for gain reduction can then be set. Ideally, for good large signal handling, gain ahead of the frequency selective stages would be reduced as early as possible, but since it is essential to maintain, or increase beyond a minimum, the signal into the ADC, this route must be modified.

The simplest implementation of an AGC function would be a floating point processor, but this would suffer from interference from unwanted signals. A better approach is to use the DSP signal level output, digitally integrated over a short time, and to apply it to the ADC as a multiplying command, either in digital form, or by an analogue method ahead of the ADC. A really large signal should bring in a switch selected front-end attenuator, but this could be relatively coarse, e.g. 20dB.

1.6.7 Control functions

Clearly, a receiver of this type needs extensive computer control. Present generation sets use a serial control line (RS 232) for remote control of some functions, but rely on an internal dedicated processor for most or all operational requirements. In principle, all of this could be carried out in a standard computer, with the radio installed as a card within the machine, with full access to the bus. Practically, at least one 'software radio' receiver (ICOM PCR-1000) exists where all front end processing, and some DSP, exists in a small external box, to avoid computer generated interference. The bulk of the signal processing and all the control is in the PC itself.

1.7 Conclusion

Digitally implemented radios for the HF bands are feasible and have been implemented by at least two suppliers (Rockwell-Collins and Watkins-Johnson). Other companies have the capability and active research groups. It is to be expected that many more products in this field will be launched in the near future.

Digital techniques for the HF bands are limited by the need for extreme dynamic ranges, and the architecture proposed and implemented to date reflect the compromises required. However, the use of digital techniques offers greater flexibility in mode selection, filter final bandwidth and in some cases fine frequency control.

For VHF and higher frequencies, software oriented receivers are already down to prices where amateur use is feasible.

Further advances in ADC design for more accuracy and wider bandwidth are needed for significant steps forward in performance to occur. On the DSP side, there is some evidence of major steps in design thinking, especially with the digital down converter concept.

Ultimately, a very small chip set could be designed to realise all functions of a digital radio; this cannot be more than a few years away.

1.8 Appendices 1 Signal Handling Considerations
 

Shading approximately indicates atmospheric noise limit.  
 

1.9 Appendices 2: DSP Processors

  After Grant, ref. 14, with additions  
 

HSP 50016 Notes
1) Strictly   0.140625 x sample rate/HDF decimation rate
2) Strictly   0.1992188 x sample rate/HDF decimation rate

[back to top]

Site Search

Powered by Google
Site Map