DESIGN CONSIDERATIONS FOR DIRECT RF SAMPLING RECEIVER IN GNSS

ENVIRONMENT

Ville Syrjälä, Mikko Valkama, Markku Renfors

Tampere University of Technology

Institute of Communications Engineering
P.O Box 553, FIN-33101 Tampere, Finland

ABSTRACT within the sampling process itself, using aliasing in a

controlled manner. So there is no need for analog
In this paper, an advanced direct RF sampling receiver mixers and thus the analog component count decreases.
architecture is studied for the GNSS environment. The Analog components are usually bulky, expensive and
architecture is based on sampling the signal directly at use quite much power so the basic concept of the DRFS
RF, which in the GNSS case are in the 1.5 GHz range. gets very interesting. In addition, the flexibility of the
The high-frequencies in the signals to be sampled pose DRFS receiver is at a very high level because the signal
then very high demands for the accuracy and quality of is sampled directly from the radio frequencies (RF).
the sampling process, and thus quantization and Naturally, the signal must be filtered before sampling
especially the timing jitter must be considered in detail. but the filtering can be done so that the signals of more
The study shows that the quantization and jitter than one communications systems can be processed and
requirements are, however, feasible when the pre- received. The topic of DRFS has not yet been
sampling filtering is done properly. thoroughly investigated and there are only a few papers
that consider the practical implementation of the DRFS
receiver principle. Thus, the idea is very promising but
more research needs to be done on the topic. The direct
1. INTRODUCTION RF sampling receiver architecture has been considered
in [1], [2] and [12] with focus on the GNSS case, and
Not only the wide scale changes in technology of more generally in [4].
Global Navigation Satellite Systems (GNSS), but also One of the main problems in a DRFS receiver is the
the demand for low-cost and flexible GNSS receivers high demands for the quality of the sampling process
has been greatly increasing in the last few years in many due to the high frequency of the sampled signals [6],
application areas. Car navigation systems are becoming [15]. This paper concentrates on the requirements the
almost an essential part of better-class cars, and the DRFS receiver has for the sampling process in GNSS
demand for GNSS integrated cellular phones is also application. First, the sampling jitter and the
increasing [5]. The car navigation systems need to be as quantization effects on the signal-to-noise ratios (SNR)
cheap as possible and cellular phone integrated are considered. Then, the sampling jitter and
navigation systems must also be small and have low quantization level requirements for the sampling of a
power consumption. In addition, as the cellular phones GNSS signal centered at 1575.42 MHz (E2-L1-E1)
of course need receivers for capturing the cellular band are considered. After the requirements have been
signals, there is big interest to receive Global System for derived by using system calculations, the results are
Mobile Communications (GSM), Universal Mobile verified using simulations.
Telecommunications System (UMTS) and GPS/Galileo
signals, all with one flexible receiver. 2. BASIC NON-IDEALITIES IN SAMPLING
Direct RF Sampling (DRFS) receiver architecture is
one promising approach for building flexible radio The main sources of noise in a sampling system are the
receivers. The idea of DRFS is to sample the signal at as limited number of bits in the quantization and
early stage of the receiver as possible. The down- unintentional deviations called jitter in the moments the
conversion of the high-frequency signals can be done samples are taken. Limits in these non-idealities are of
interest when a high-performance sampling system is telecommunications, we can assume for analysis
needed in demanding environments such as in DRFS purposes that the interesting signal is a sinusoidal signal
receiver [13]. For system calculations, we focus on as well. Thus, its voltage can be presented as
power spectral densities (PSD) of the noise components V (t ) = A cos(2π f 0 t ) . With this assumption, we are able
generated in the sampling process. In this chapter, to calculate the voltage error of the signal when we
PSD’s for quantization and jitter noise are determined. consider only the error due to the jitter. In the jitter
These are then used in Chapter 3 for deriving the analysis, we use the RMS value of the timing jitter. Let
quantization and jitter requirements for capturing the us first calculate the RMS value of the derivate of V:
GNSS signal using the DRFS principle.
T0 2
δV 1 ⎛ δ V (t ) ⎞
T ∫0 ⎝ δ t ⎠
= ⎜ ⎟ = 2 Aπ f 0
2.1. Quantization Noise δ t rms

The most basic non-ideality in a sampling system is the In this equation, T0 = 1/ f 0 is the cycle time of the
quantization noise. If a sampled signal has values from sinusoid. When we multiply the RMS value of the
−1 to 1 and n bit quantization is used, the possible error voltage derivative by the RMS value of the jitter Δt , we
value is uniformly distributed between -2- n and 2- n . get the voltage error, which is 2Aπ f 0 Δt . Then, by
Thus the power of the quantization noise is dividing the signal power by jitter noise power, the SNR
due to jitter is of the form:
2−2 n
var(e(i )) =
(A/ 2)
2
3
SNR jitter = = (2π f 0 Δt ) −2
( )
2
where e(i) is the error of the i-th sample. Assuming a 2 A π f 0 Δt
sinusoidal signal with root mean square (RMS) power
of (Vrms ) 2 = 1/ 2 , the SNR due to quantization can be In [13], Shinagawa et al have done some calculations
written as: with the same results and noticed that this model is as
1/ 2 3 accurate as expected. Now that the SNR jitter is
SNRquant = = 22 n calculated and we have got rid of the amplitude term,
2 −2 n / 3 2
the power of the jitter Pjitter can be calculated as
In quantization, the white noise model is considered to
Pjitter = Ptotal / SNR jitter
be valid and thus the PSD of the quantization noise can
be written as: Then if the noise is considered to be white noise and
Pquant Psignal 2 Ppeak thus spread over the Nyquist band, we can get the PSD
PSDquant ( f ) = = = (1) of the jitter noise as
Fs SNRquant Fs 3Fs 22 n
Pjitter Ptotal (2π f 0 Δt ) 2
In the final form, we have substituted Psignal = Ppeak PSD jitter ( f ) = = (2)
Fs Fs
since the peak power is the critical quantity in
quantization. White noise naturally falls to the Similar results are got by Valkama et al. in [16].
bandwidth determined by the sampling rate Fs .
The calculations of quantization noise are based on 3. JITTER REQUIREMENTS IN DIRECT RF
the basic signal processing work done by Huttunen in SAMPLING RECEIVER
[10], and by Tsui in [14].
One possible approach to determining the jitter
2.2. Jitter Noise requirements in a GNSS receiver was introduced by
Dempster in [6]. The approach is very simple and is
As the frequencies of the sampled signals rise, the based on the basic properties of the GNSS signals. As is
timing of the sampler becomes a limiting factor in the known, the GNSS signals are under the thermal noise
performance of the sampling process [3], [13]. In this level. GPS signal is not expected to exceed −150 dBW
section, we study the contribution the noise generated in any situation [11] and is thus under the thermal noise
by the timing jitter has on the performance of the total with its about 2 MHz bandwidth. Also, Galileo signal
system. specifications say that its signals in E2-L1-E1 band are
The jitter noise naturally depends on the shape of not expected to exceed −148 dBW [8], and have wider
the signal we are sampling. Because the transmitted bandwidth of interest than signals of the GPS. The
modulated signals are of oscillating nature in
 1
PSD jitter ( f ) ≤ PSDth ( f ) (3)
10
Thus, the jitter requirement can be written, with help of
Equation (2) and Equation (3), in the following way:

Fs
Δt ≤
20 × B (1 + x)(2π f 0 ) 2

Now by denoting the oversampling factor osf = Fs / B

in the equation, the simplest form can be achieved
without the bandwidth term:
Figure 1: The noise-dominated environment. The
signal with thicker line is the signal we are osf
Δt ≤ (4)
interested in. 20 × (1 + x)(2π f 0 ) 2
Galileo signals are also under the thermal noise. Thus, By using Equation (4), we get the basic results for the
when sampling a GNSS signal at RF, the sampled signal jitter requirements. The results are here calculated for
practically consists of the thermal noise or other oversampling factors of four and eight as thought
interferences that are not filtered out. As was shown in realistic in band-pass sampling. These correspond to
Equation (2), the power of the jitter noise depends on sampling rates of 120 MHz and 240 MHz with signal
the overall power of the sampled signal. Now, if the bandwidth of 30 MHz, which should be enough for
noise generated by the jitter is say 10 dB under the Galileo reception [1], [9]. Calculating the results for
thermal noise, its contribution to the total system noise more oversampling factors is a trivial process.
can be considered insignificant. The situation explained As the results in Table 1 show, if there is no
here can be seen in Figure 1. The thermal noise is at a external interference in the system, the requirements are
higher level than the interesting signals and interfering not so tight at all, but when the interference increases,
in-band signals. In addition, some out-of-band the requirements get tighter and tighter. Novel low-
interference signals may be present. power analog-to-digital converters and samplers have
Although ideally the signal we take samples from is RMS jitter value of about two picoseconds [4].
thermal noise, there might be other interfering signals Therefore, the tightest requirement with oversampling
present if our pre-sampling filtering is not selective factor of eight is still achievable but that would be quite
enough. In addition, stronger other GNSS signal might tightly in the limits. In interference-free environment,
interfere with the current signal of interest. Therefore, in the sampling jitter requirements would not be hard to
determining the overall power of the sampled signal, achieve, but getting such a semi-ideal environment
these other signals need to be taken into account. There would result in very high requirements for the
is thus interest to study cases where the total sampled selectivity of the front-end filters. With more power-
power is, for example, 10 dB, 20 dB and 30 dB over the consuming analog-to-digital converters and samplers,
thermal noise, to make sure the study is surely extensive all the calculated requirements are easily achievable, but
enough. This means that the total sampled power Ptotal the practical implementation in mobile receivers could
can be written as: not be done with the technology of today.
Ptotal = N th (1 + x) = N th + xN th
4. QUANTIZATION REQUIREMENTS IN
in which x represents how many times stronger the DIRECT RF SAMPLING RECEIVER
overall out-of-band interference is, compared to the
thermal noise. For 10 dB, 20 dB and 30 dB the values of The quantization requirements can in principle be
x are 10, 100 and 1000 respectively. The PSD of the calculated in a rather similar way as the jitter
thermal noise itself is [7]: requirements. If the quantization noise is 10 dB under
the thermal noise, the contribution is no more than that
1 of the jitter noise in the same situation. As the Equation
PSDth ( f ) = kT
2 (1) shows, the quantization noise power depends
The jitter requirements for voltage sampling can then be essentially on the peak power of the sampled signal.
derived from the equation: Thus, the difference between the jitter and the
quantization study is that as the jitter noise depends on
 Table 1: The jitter requirements for voltage Table 2: The quantization requirements for DRFS
sampling DRFS receiver receiver.
Oversampling factor Oversampling factor
Ptotal / N th Ptotal / N th
4 8 4 8
0 dB 45.2 ps 63.9 ps 0 dB 2.53 (3) bits 2.03 (3) bits
10 dB 13.6 ps 19.3 ps 10 dB 4.26 (5) bits 3.76 (4) bits
20 dB 4.50 ps 6.36 ps 20 dB 5.86 (6) bits 5.36 (6) bits
30 dB 1.43 ps 2.02 ps 30 dB 7.51 (8) bits 7.01 (8) bits

the total sampled power, the quantization noise depends

on the peak power of the signal. This is natural because decimated without filtering to get a sampling frequency
the highest signal amplitude must have its own level in of 237.5 MHz. This is equivalent for the oversampling
the quantization process. We also know that in the factor of eight for a bandwidth of 29.6875 MHz. Due to
GNSS environment, we practically sample band-pass the decimation by 16 without filtering, the jitter noise
noise whose crest factor is around 10 (peak / RMS). We aliases to lower frequencies as it would have aliased if
can thus write: the 237.5 MHz sampling frequency would have been
used in the first place. The spectrum at the simulator
Ptotal = 1/10 × Ppeak output can be seen in Figure 2.
The simulations verify the fact that the jitter noise
Putting all these together the requirement for the stays 10 dB under the thermal noise with 63.9 ps
quantization accuracy can be formulated as sampling jitter in case of oversampling factor of eight.
⎛ 400 × (1 + x) ⎞ In addition, let us verify one more case with
n ≥ log 2 ⎜ ⎟/2 (4) oversampling factor of 4. This is equivalent with
⎝ 3 × osf ⎠ decimation by 32. Thus the sampling rate is 118.75
The requirements can be directly evaluated from the MHz. In this case, the sampling jitter of 45.2 ps should
Equation (4). The results can be seen in Table 2. take the jitter noise 10 dB under the interesting signal.
As the results in the table demonstrate, the The results are shown in Figure 3. It can be seen that the
quantization requirements are from two to eight bits jitter noise is, indeed, 10 dB under the band-pass-noise
depending on the case. The requirements are not so high signal and thus the jitter requirements seem to be valid
and are achievable with novel low-power analog-to- for both oversampling factors used in calculations.
digital converters. The quantization noise does not act
as strong limiting factor in a DRFS receiver as the 5. CONCLUSIONS
sampling jitter noise does.
We have studied the jitter noise and quantization noise
5. JITTER REQUIREMENT VERIFICATION effects in the sampling and A/D process with special
focus on direct RF sampling receiver in GNSS context,
Next, our target is to verify by simulations the jitter and derived the equations for power spectral densities of
calculations we made in the previous sections. The fact such noises. Furthermore, the equations for jitter and
that the jitter requirements were obtained only by using quantization requirements for direct RF sampling
the system calculation principles, leaves the results receiver in GNSS environment were derived. It was
floating a little. Even though the used equations are assumed that if the power of the sampling noise is 10
verified with simulation in the references, the results we dB under the thermal noise, the contribution to the total
got should also be verified. As we saw in the results in system performance is diminishing. The jitter and bit-
Table 1, 63.9 ps jitter is enough for us to get the jitter resolution requirements were considered in a few
power 10 dB under the sampled pass-band noise signal realistic situations. The results showed that if the
power if oversampling factor of 8 is used. Now we will selectivity in the RF filters is good enough, the bit and
verify this result. jitter requirements are feasible. On the other hand, if the
We make a simulator with MathWorks Matlab 7. In pre-sampling filtering is not done properly, the
the simulator, a band-pass-noise signal with 29.6875 requirements for the jitter rise to levels that are not yet
MHz bandwidth at E2-L1-E1 band (1575.42 MHz practical in mobile communications devices, at least
centered) is sampled. We use a 3.8 GHz sampling rate with today’s techniques. The jitters of around, or less
in the actual jitter simulation stage and the signal is then than, one picoseconds are achievable with high-power
analog-to-digital converters of today but are by no 6. REFERENCES
means practical in mobile devices due to high power
consumption. The resolution requirements seemed quite [1] Akos D. M., Ene A., and Thor J, “A Prototyping
relaxed even with high interfering signal levels. Platform for Multi-Frequency GNSS Receivers,” in
Practically, eight bits would be enough even in very Institute of Navigation GPS/GNSS Meeting, Portland,
2003.
demanding applications.
Simulations verified that with the calculated [2] Akos D. M., and Tsui J. B. Y., “Design and
sampling jitter requirements, the jitter noise stayed 10 Implementation of a Direct Digitization GPS Receiver
dB under the thermal noise, which was used as the Front End,” IEE Transactions on Microwave Theory and
design principle in jitter calculations. Techniques, Vol. 44, No. 12, December 1996.

[3] Amin B., and Dempster A. G., “Sampling and Jitter

Considerationg for GNSS Software Radio Receivers,” in
IGNSS Symposium 2006. 17 – 21 July 2006.

[4] Arkensteinjn V. J., Klumperink E. A. M., and Nauta B.,

“Jitter Requirements of the Sampling Clock in Software
−35 Radio Receivers,” IEEE Transactions on Circuits and
Systems – II: Express Briefs, Vol. 53, No. 2, February
−40 2006.
Power [dBW]

[5] Behbahani F., Firouzkouhi H., Chokkalingam R.,

−45
Delshadpour S., Kheirkhani A., Nariman M., Conta M.,
and Bhatia S., “A Fully Integrated Low-IF CMOS GPS
−50 Radio with On-Chip Analog Image Rejection,” IEEE
Journal of Solid-State Circuits, Vol. 37, No. 12,
−55 December 2002.

−100 −50 0 50 100 [6] Dempster A. G. “Aperture Jitter Effects in Software

Frequency [MHz] Radio GNSS Receivers,” in The 2004 International
Symposium on GNSS/GPS, GNSS 2004, December 6-8,
2004.
Figure 2: Band-pass-noise signal sampled with
237.5 MHz sampling rate with 63.9 ps sampling [7] Egan W. F., Practical RF System Design, John Wiley &
jitter. Sons, Inc. 2003. ISBN 0-471-20023-9. 386 p.

[8] European Space Agency (ESA) / Galileo Joint

Undertaking, “Galileo Open Service in Space Interface
Control Document (OS SIS ICD) Draft 0.” 2006.
−40
[9] Heinrichs G., Bischoff R., and Hesse T., “Receiver
−45 Architecture Synergies between Future GPS/Galileo and
Power [dBW]

UMTS/IMT-2000,” IEEE 2002.

−50 [10] Huttunen H., Signaalinkäsittelyn sovellukset, Tampere
University of Technology, Institute of Signal Processing,
−55 2006. ISBN 952-15-1542-2, ISSN 1459-4609. 117 p.

−60 [11] IS-GPS-200. “Interface Specification – Navstar GPS

Space Segment/Navigation User Interfaces,” IS-GPS-
−50 −25 0 25 50 200D, December 7, 2004.
Frequency [MHz]
[12] Psiaki M. L., Akos D. M., and Thor J., “A Comparison
of ”Direct RF Sampling” and ”Down-Convert &
Figure 3: Band-pass-noise signal sampled with Sampling” GNSS Receiver Architectures,” in ION
118.75 MHz sampling rate with 45.2 ps sampling GPS/GNSS 2003, September 9-12, 2003.
jitter.
[13] Shinagawa M., Akazawa Y., and Wakimoto T., “Jitter
Analysis of High-Speed Sampling Systems,” IEEE
 Journal of Solid-State Circuits, Vol. 25, No. 1, February
1990.

[14] Tsui J., Digital Techniques for Wideband Receivers,

Artech House, 2001. ISBN 1-58053-299-3. 588 p.

[15] Walden R. H., “Analog-to-Digital Converter Survey and

Analysis,” IEEE Journal on Selected Areas in
Communications, Vol. 17, No. 4, April 1999.

[16] Valkama M., Pirskanen J., Renfors M., “Signal

processing challenges for Applying Software Radio
principles in Future Wireless Terminals: An Overview,”
International Journal of Communication Systems, 2002.
15:741-769.