Professional Documents
Culture Documents
1 O=1
Normalized amplitude gain [−]
O=2
0.8 O=3
O=4
0.6
0.4
Fig. 3. Trade-off between frequency and resolution
0.2
does, to improve linearity, is to move the upper frequency break-
0 1 2 3 4 5 6 7 8 point by advances in technology (and clever design tricks). This upper
Frequency normalized w.r.t. sample frequency [−] frequency break-point is partly given by the transition-frequency, fT ,
of the transistor. There are however other parameters that play a role
Fig. 2. Frequency characteristics for different return-to-zero schemes too, and one needs to find the optimum values for the application. By
knowing these limits we also know what barriers we cannot breach.
In the favorable current-steering DAC, the current sources will
high-frequency components (i.e. images of baseband spectrum), such have a limited output impedance. For a linear system, this would
that we can use other Nyquist ranges than that at frequencies below not cause us any major issues, it would just result in a gain error
half the sample frequency. in the output. There would “only” be a loss of signal power due to
the unmatched impedance levels. However, as the output impedance
A. Frequency-domain aspects
varies with signal level, it affects the linearity significantly. The output
In terms of how the DACs behave in the frequency domain, we voltage is dependent on the code as
refer to Nyquist-rate, interpolating, and oversampling converters.
ZL
The Nyquist-rate converter is designed to use the entire available Vout (X) = iout (X) · (4)
signal band from ωT = 0 to ωT = π, (i.e., up to the Nyquist 1 + ZL /ZS (X)
frequency). Quite often this is an over-design in many applications, where ZL is the load impedance and ZS is the output impedance
since the signal actually can be quite narrow banded. of the DAC which is in the unbuffered current-steering DAC also
There are two flavors of interpolating DACs: analog [6] and digital a function of input code, X. Each current source will have a unit
interpolation. Both share the property of not being able to reproduce output impedance, Z0 , and if the corresponding weight is 2n , the
the entire Nyquist band. effective output impedance will be Zunit /2n . We can generalize this
Even higher linearity can be reached by further employing the and say that if the code X is applied to the DAC there will be
interpolation technique and also introducing the sigma-delta mod- X current sources connected to the output and hence the effective
ulator. For DACs, sigma-delta it is somewhat more attractive than output impedance will be Zs = Zunit /X. This means that the output
for ADCs, since the modulators in the DAC can be implemented in voltage can be written as:
digital domain. iunit · X · ZL ∆V · X
Vout = = (5)
B. A comment on high-speed DAC architectures 1 + ZZ L
unit
· X 1 + X/ρ
The most suitable architecture for high-speed DACs is the current- where we have set ∆V = iunit ·ZL as our least significant bit (LSB)
steering architecture as it has the benefits of not (necessarily) requir- voltage step at the output of the DAC and ρ = Zunit /ZL is the ratio
ing an output buffer for high performance. It directs almost all of its between output and input impedance of a unit source. The equation
current to the output, which means high efficiency. The current can above is nonlinear with respect to the input code X and distortion
also be terminated off-chip to minimize on-chip heat dissipation. The will be introduced: the harmonic distortion (HD) will depend on the
weighting elements of the DAC are implemented as current source impedance ratio and the amplitude. The peak amplitude, X0 , is given
and switches and we can typically write the (static) output current by the number of bits in the DAC as X0 ≈ 2N . The full-scale output
as: current is iunit · X0 , and quite often we have levels around iunit ·
iout (X(n)) = iunit · X(n) (3) X0 · ZL ≈ 1 V. The ρ value is quite likely large, but for a higher
number of bits, the X0 grows larger too. The problem is also that the
for the code X(n) and iunit is the output current of a unit current ρ decreases steadily with higher frequency. It can be shown [4] that
source. However, there are several significant limitations on achiev- the third order distortion for a differential DAC can be approximated
able speed and performance in current-steering DACs too. as
III. N ON - IDEAL DAC S HD3 = 40 · log10 ρ − 12 · (N − 2). (6)
As illustrated in Fig. 3: there is normally a relationship between So, for an output impedance of 100 MΩ, the harmonic distortion
resolution and frequency in digital-to-analog converters [4], [7], [8]: becomes some 45 dB for a 14-bit DAC. For a 12-bit DAC it becomes
with higher frequencies, the linearity decreases. There is not that 57 dB. If the output impedance is some 100 MΩ at 10 kHz and the
much you can do about the overall tendency, but there are, of load is 50 Ω, we get HD3 ≈ 110 dB. Assuming a slope of 20 dB per
course, different sources for limited performance, and they might decade, we will at 1 MHz have 1 MΩ output impedance, resulting in
have different slopes and one could design the DAC to track the HD3 ≈ 70 dB, etc. Major challenges are therefore to increase output
best of these slopes over the frequency domain. What one generally impedance of the converter and/or to lower its load impedance.
In the right-most part of Fig. 5 we find an active implementation
of the mixer DAC. In this case we have actually illustrated this with a
single unit element of the DAC. Then, several of these units must be
connected in parallel to achieve the overall functionality of the DAC.
We know since previously that the static output current is given by
where, for example, X(n) = X0 ·sin(ω1 T n), could contain the signal
information and p(t) could be the brickwall function or a return-to-
zero. Now, assume that iunit instead is a time-dependent oscillating
signal as
iunit (t) = iunit,0 · q(t) (9)
In this section we outline some different architectures from a If q(t) now is given by the local oscillator (LO) signal, sin(ω0 T n),
system-level point of view. The top of Fig. 4 outlines a traditional we see that the mixed products at ω0 ± ω1 are produced such that
IQ transmitter with the DC/low-IF baseband digital processing, the we can center the signal information at higher frequencies. One
digital-to-analog converter (DAC) that generates the analog waveform problematic design issue is to distribute the mixing signal to multiple
which is then bandpass or low-pass filtered (BPF/LPF), then mixed, unit source and maintaining matching and skew.
summed, filtered again (BPF) and brought to the antenna by the power
amplifier (PA). This is of course a simplified picture, but illustrative
for our purpose here. B. Interpolating DAC
The idea with the RF DAC or mixer DAC is to combine the mixer In literature you sometimes find the interpolating DAC as a high-
with the DAC in a clever way i.e. by reducing the number of analog speed converter. The digital interpolating DAC is however nothing
components. We illustrate this in Fig. 4 where parts of the baseband, else but a normal, Nyquist-rate DAC combined with a digital interpo-
the DAC, filters, mixers and optionally the final bandpass filter can be lator: we apply digital multi-rate interpolators at the input of the DAC
merged. One result of this operation is that the digital sampling speed, such that the update frequency is increased and signal can be located
as well as the requirements on linearity and frequency selectivity at higher frequencies by applying digital filtering. The advantage is
increase quite dramatically. that we can feed the DAC with a lower frequency input thus relaxing
the interface to the circuit. In the case of analog interpolation we try
A. DAC and mixer instead to mimic the desired analog output waveforms by for example
The most intuitive approach is to combine the DAC directly with microstepping methods as in e.g. [6] or by using return-to-zero
a mixer. This will constitute a traditional way of modulating and schemes which effectively introduce similar kind of zero-padding as
transmitting. The ambition should be to push the DAC closer to the the digital interpolation does, but with maintained requirements on
antenna by preferably increasing its sampling frequency as much as the analog filters.
possible. We also want to push the mixer closer to the DAC and The return-to-zero scheme, i.e., using a PAM waveform which is
such that we can remove the resistive load at the output of the DAC. unity for 0 < t < T /2 rather than 0 < t < T , as effectively gives
The left-most part of Fig. 5 shows an example of a DAC element you zero padding and a form of interpolation. The inverse clock
merged with a passive mixer (passive in the sense that there is no signal can be used to “easily” create the zero padding. Notice, that
gain associated with it). For this purpose we integrate the mixer close the PAM waveform is still defined such that it has the same sample
to the DAC and without the resistive load. The current-to-voltage-to- frequency as the brickwall one. Advantages with the return-to-zero
current conversion – which introduces distortion – is avoided. The scheme is that we reduce both analog settling memory effects, as well
load impedance is also reduced by sinking the DAC output currents as memory effects inside the digital logic, thus improving linearity
in the low-impedance drains of the mixer transistors. and noise.
Fig. 7. Simplistic view of a sigma-delta modulator
R EFERENCES
[1] A. Majid and A. W. Malik, “Design and implementation of a direct
digital frequency synthesizer using sum of weighted bit products,”
Master’s thesis, Linköping University, Sweden, 2009.
Fig. 8. DAC signal chain used by Luschas, et al. [11] [2] S. Halder, H. Gustat, and C. Scheytt, “A 20GS/s 8-bit current steering
DAC in 0.25µm SiGe BiCMOS Technology,” in Proc. European
Microwave Integrated Circuits Conf., EuMIC, 2008, pp. 147–150.
With a lower number of analog components we can design for [3] J. Hägglund, “Pulse and noise shaping D/A converter (PANDA) Block
implementation in 65nm SOI CMOS,” Master’s thesis, Linköping Uni-
high-speed and apply dynamic element matching (DEM) techniques.
versity, Sweden, 2008.
These techniques use randomization to cancel out signal-dependent [4] J. J. Wikner, “Studies on CMOS Digital-to-Analog Converters,” Ph.D.
components and transforming this energy into noise. We have pre- dissertation, Linköping University, Sweden, Apr. 2000, Dissertation No:
viously shown methods that can be applied to spectrally shape the 667.
error energy such that it falls out of band and does not degrade the [5] K. O. Andersson, “Modeling and implementation of Current-Steering
Digital-to-Analog Converters,” Ph.D. dissertation, Linköping University,
effective signal-to-noise ratio (SNR) [15]. Sweden, May 2005, Dissertation No: 944.
[6] Y. Zhou and J. Yuan, “A 10-bit 100-Mhz CMOS linear interpolation
V. C OMPARISON AND C ONCLUSIONS DAC,” in Proc. European Solid-State Circuits Conf., May 2002, pp.
471–474.
Some interesting results on RF DACs were presented in [11] and [7] S. Luschas and H.-S. Lee, “Output impedance requirements for DACs,”
Fig. 8 outlines the basics of that architecture. In fact, it is rather in Proc. IEEE Intern’l Conf. Electronics, Circuits and Systems (ICECS).,
straight-forward approach: a sigma-delta modulator used to reduce Sep 1999, pp. 1193–6.
[8] A. V. den Bosch, M. Borremans, M. Steyaert, and W. Sansen, “A 10-bit
the number of analog components in the RF DAC. Dynamic element
1-GS/s Nyquist current-steering CMOS D/A Converter,” IEEE J. Solid-
matching (DEM) is used to scramble data and decorrelate static State Circuits, vol. 36, no. 3, Mar. 2001.
errors from the signal to further reduce nonlinearity. The sigma- [9] R. Carver, “Double DAC rate by using mixers as switches,” in EDN,
delta modulator is an effective way to spectrally shape quantization Feb 2004, pp. 69–70.
errors such that a DAC with less analog contents can be used. [10] S. Park, G. Kim, S.-C. Park, and W. Kim, “A Digital-to-Analog
Converter based on differential-quad switching,” IEEE J. Solid-State
The results they present are elegant in the way that they have Circuits, vol. 37, no. 10, Oct. 2002.
successfully implemented the mixing switching scheme to the DAC [11] S. Luschas, R. Schreier, and H.-S. Lee, “Radio Frequency Digital-to-
together with sigma-delta converters. They have also demonstrated Analog Converter,” IEEE J. Solid-State Circuits, vol. 39, no. 9, Sept.
that the oscillator signal connected to the internal elements of the 2004.
[12] S. Noli, A. P. Perez, E. Bonizzoni, and F. Maloberti, “∆Σ time
unit DAC elements does not have to be that accurate to achieve interleaved current steering DAC with dynamic elements matching,” in
promising results. They have chosen waveforms such that the spectral Proc. IEEE Intern’l Symp. Circuits and Systems., 2009, pp. 407–10.
attenuation of the DAC output is not that high at certain frequency [13] J. Pham and A. C. Carusone, “A time-interleaved ∆Σ-DAC architecture
bands above the sample frequency. (See Fig. 2 for comparison). clocked at the nyquist rate,” IEEE Trans. Circuits Syst. II, vol. 55, no. 9,
Instead, the lobes at higher frequency are amplified compared to the Sept. 2008.
[14] A. Jerng and C. G. Sodini, “A wideband ∆Σ Digital-RF modulator
traditional sinc waveform, such that a local maximum is achieved with self-tuned RF bandpass reconstruction filter,” in Proc. IEEE Custom
at approximately twice the sample frequency, equal to the oscillating Integrated Circuits Conf., CICC, 2006, pp. 125–8.
frequency. They report 75 dBc linearity in a 17.5-MHz band centered [15] M. Vesterbacka, M. Rudberg, J. J. Wikner, and N. U. Andersson, “Dy-
at 942 MHz. The technique is promising, but it is not wide-band namic element matching in D/A converters with restricted scrambling,”
in Proc. IEEE Intern’l Conf. Circuits and Systems (ICECS 2000), 2000,
enough. It must also be mentioned that for example digital pre- pp. 36–39.
distortion (DPD) required to linearize e.g. the power amplifier will [16] S. M. Taleie, T. Copani, B. Bakkaloglu, and S. Kiaei, “A linear Σ∆
need a bandwidth a couple of times wider than the signal band for Digital-IF to RF DAC transimtter with embedded mixer,” IEEE Trans.
proper cancellation of harmonics. To not destroy the properties of the Microwave Theory Tech., vol. 56, no. 5, May 2008.
[17] P. Eloranta and P. Seppinen, “Direct-Digital RF modulator IC in 0.13µm
DPD we cannot narrow down our bandwidth through the DAC too
CMOS for Wide-Band Multi-Radio applications,” in Proc. IEEE Intern’l
much. Table I summarizes the performance of different architectures Solid-State Circuits Conf. (ISSCC), Feb 2005, pp. 532–4.
of high speed DAC, reported in the recent publications. Also included [18] B. Schafferer and R. Adams, “A 3v cmos 400mw 14b 1.4gs/s DAC for
in the table are some commercial products from Maxim and Analog multi-carrier applications,” in Proc. IEEE Intern’l Solid-State Circuits
Devices. Conf., Feb 2004, pp. 69–70.