SlideShare a Scribd company logo
1 of 4
Download to read offline
Study of Modified Noise-Shaper Architectures
           for Oversampled Sigma-Delta DACs
                          Nadeem Afzal                                                    J Jacob Wikner
              Department of Electrical Engineering,                           Department of Electrical Engineering,
      Link¨ ping University, SE-581 83 Link¨ ping, Sweden
          o                                 o                         Link¨ ping University, SE-581 83 Link¨ ping, Sweden
                                                                          o                                 o
                E-mail: nadeem@isy.liu.se                                      E-mail: Jacob.Wikner@liu.se



   Abstract—In this paper, modified low-complex, hybrid archi-        results are found in Section IV. Finally, we conclude the results
tectures for digital, oversampled sigma-delta digital-to-analog      and elaborate on some future extensions to the work.
converters (Σ∆DACs) are explored in terms of signal-to-noise            In this paper, further exploration and modification of the
ratio (SNR) and subDAC complexity. The studied techniques
illustrate the trade-off in terms of noise-shaper and DAC im-        work in [3] is done and some novel results are presented.
plementation complexity and loss in SNR. It is found that a fair
amount of improvement in SNR is achieved by maintaining low-                        II. S IGMA D ELTA M ODULATOR
complexity of noise shaper. The complexity of the subDAC is yet         A simplified interpolation chain for digital-to-analog con-
a parameter, directly related to the number of output bits from      version including a noise-shaping Σ∆ modulator is shown in
the noise shaper.
   Two different architectures are investigated with respect to      Fig. 1. We use a slightly unorthodox starting point to describe
subDAC complexity and noise shaper complexity. It is shown           the theoretically achievable SNR in the signal path if the the
that the required number of DAC unit elements (DUE) can be           noise-shaping module is omitted:
reduced to half.
   Index Terms—Sigma-Delta-Modulator, Noise Shaper, DAC-                                   SNR = log2 (DUE),                       (1)
Complexity, Modulator’s Complexity, Hybrid architecture.
                                                                     where DUE is the required number of unit elements in the
                      I. I NTRODUCTION                               DAC [7]. For the “standard” Nyquist converter, i.e., interpo-
                                                                     lation factor is 1 in the signal chain), DUE would simply be
   The sigma-delta modulator (Σ∆), among many other mod-             given by
ulators, attracts remarkable appreciation mainly due to the po-                                DUE = 2N ,                      (2)
tential to reach high accuracy, i.e., high resolution. It achieves
high signal-to-noise ratio (SNR) by utilizing a high oversam-        and the obtainable (theoretically) SNR at the input of the DAC
pling ratio (OSR), where OSR is defined as the ratio between          would be
the sampling frequency and the baseband signal bandwidth.                             SNR = 6.02 · N + 1.76 dB.                  (3)
To keep OSR constant for high bandwidths applications, the
sampling frequency naturally has to be increased. An increased          So, in short: for a Nyquist rate converter, to achieve high
sampling frequency is restricted, not only by the mismatches         SNR, a high number of bits, N , is required implying a very
because of the process technologies but also by the increase         complex DAC for high-speed applications. In the case of a
in complexity of hardware circuitry.                                 limited bandwidth compared to the Nyquist frequency, i.e.,
   In cases of wideband signals and/or lower OSR applications,       oversampling is possible, the SNR can be increased by a factor
the SNR can be improved by increasing the number of output           10 · log10 (OSR) [8] implying that
bits of the noise shaper, i.e., the number of quantizer bits.              SNR = 6.02 · N + 1.67 + 10 · log10 (OSR) dB.            (4)
This is done at the expense of precision component mismatch
in the output digital-to-analog converter (DAC) [1]. One             However, just utilizing a high OSR is not always possible
can also increase the order of the Σ∆ to further achieve             since, still, high-speed architectures are required. Therefore the
better performance, but the increment in the order makes the         number of bits fed to the DAC is reduced by employing the
large domain of parameters difficult to be optimized. Hence,          noise shaper. It can effectively be used to maintain the same
a higher-order Σ∆ is more difficult to design in terms of             SNR within the signal band eventhough the number of bits
stability [2]. Conclusively, in the low-OSR scenario, high           is lower. The number of bits is reduced by re-quantization at
resolution is difficult to reach with a single-bit quantizer -        coarser levels while the additional quantization noise is shifted
even with the high-order Σ∆ [4].                                     to frequencies outside the effective signal band. Using an ideal
                                                                     Σ∆ the in-band SNR would be formulated as [9]
  In Section II we describe some of the mathematical theory                                         (2L + 1)(OSR)2L+1
behind the obtainable Σ∆ modulators. The hybrid noise-               SNR = 6.02 · NQ + 10 · log10                     + 1.76 dB,
                                                                                                           π 2L
shaper architecture is described in Section III and simulation                                                                (5)


978-1-4244-8971-8/10$26.00 c 2010 IEEE
part XM from the input XN . In other words, XM consists
                                                                            the of MSBs of XN , whereas, XL consists of the LSBs of
                                                 Digital     Analog         XN . The output of the modulator is multiplied by a constant
                                                                            coefficient, a, and then added to the XM which was passed

                ↑
 N                                          NQ                              through unaffected. The output bits of this adder can now be
           OSR                    ∑                        DAC              fed to the DAC. The second-order Σ∆, in this architecture
                                                                            A1, is a so-called signal feed-back architecture with uniform
                                                                            quantization, as given in [8]. Since the LSBs are processed
                                                                            through the modulator, a right-shift register is employed at its
                                                                            output to match the bit positions of XM and XQ . The relation
                                                                            between the input and the output is given as [3]
      +     +    z-1          +   +
  -                       -           z-1                                                                     Y = XM + aXQ ,                             (6)

                                                                            where, XM is the signal fed forward at the final adder, a has
                                                                            the scalar value a ≤ 1. XQ is the output of the modulator,
 Fig. 1.   A simplified interpolation chain from digital to analog domain.   which in the z domain can be expressed as
                                                                                                                                2        qn
                                                                                             XQ = XL · z −1 + 1 − z −1              ·       ,            (7)
                                                                                                                                        2XM
where NQ is the number of bits in the coarse quantizer
and L is the order of the Σ∆. The gradient of shift in                      where qn /2XM is the new quantization noise being pushed
quantization noise with respect to frequency is dependent upon              towards higher frequencies [3]. Now, from the relations given
the modulator order such that the higher the order the steeper              in (6) and (7), the signal at the output is
the slope [10]. Now, we can utilize the L and NQ parameters to
achieve an SNR from (5) matching that of (4) where NQ < N .                                        XB = (XM + a · XL ) · z −1 .                          (8)
   However, we must also be careful to not increase the com-
                                                                            From (7) it is concluded that coefficient a reduces the am-
plexity of the Σ∆ too much. For very high-speed applications
                                                                            plitude of XL and qn /2XM , whereas the XM is unaffected.
the power consumption will be very high and obtainable
                                                                            For very small values of XL (compared to those of XM ),
maximum speed of the modulator might be limited by word-
                                                                            the quantization noise is further reduced by a while the
length, etc.
                                                                            signal power mainly is preserved by XM . This signal power
   In short, the output number of bits needs to be low, the
                                                                            preservation combined with the reduction of quantization noise
order should be low and also the word-length in adders and
                                                                            thus increases SNR. In the case when XM is not sufficiently
multipliers must be kept at a minimum. Naturally, there is
                                                                            larger than XL , coefficient a not only reduces the quantization
a delicate relation between all these parameters such that
                                                                            noise but also the major part of the signal power. We can
a reasonably high SNR can be obtained at reasonably high
                                                                            conclude that coefficient a must be selected according to the
hardware and power costs.
                                                                            following relation:
                       III. H YBRID A RCHITECTURE
                                                                                                               1       if NL ≥ k
                                                                                                    a=
   A hybrid Σ∆ modulator architecture, as presented in for                                                     1/2NM   if NL < k,
example [3], reduces the number of bits in the path to the
actual noise shaper, but still achieves fairly good performance.            where, NL and NM are the word lengths of XL and XM ,
There is no impact on the stability analysis, since the structure           respectively. The factor k, is chosen dependent on the strength
introduces a feed-forward path rather then feed-back path.                  of XL and the number of quantizer bits XQ .
In this architecture, the number of input bits to the noise
shaper is reduced at the cost of some additional in-band
quantization noise. In this way the architecture in [3] decreases           Input   XN                            XM            Y        DAC    Output
the complexity of the modulator, adders and multipliers be-
                                                                                                  truncator                 +             E

come smaller. However, more bits must be allocated to the                                     -
mixed-signal DAC at the output. We simply trade the digital                              +                                  a
complexity against the analog complexity. By investigating the                                      XL                 XQ
modified architecture a nice set of trade-off curves is obtained
                                                                                                                 ∑
illustrating these complexity trade-offs, both in terms of SNR
and DAC complexity.
   The modified architecture, denoted in this report as the A1
architecture, is illustrated in Fig. 2. The modulator takes XL              Fig. 2. The A1 architecture, as in [3] with reduced complexity of modulator.
as input which is obtained by subtracting the LSBs truncated
A. Sub DAC                                                                  empirically selected as
                                                                                                        
    At the output of the interpolation chain, the bits from the                                         11 if XQ = 1
                                                                                                        
 noise shaper are fed to the DAC. For high-speed applications                                        k = 9 if XQ = 2
 the current-steering DAC architecture is chosen since it pro-                                          
                                                                                                          7 if XQ = 3.
                                                                                                        
 vide high efficiency and can be implemented for high speed.
 Typically, the current-steering DACs are composed of unit                      The simulation results show that for less complex modula-
 current sources where every current source corresponds to one               tors, higher SNR can be achieved. The improvement of course
 of the 2(∗) levels, where “(∗)” is the number of input bits to              comes at some cost, which is the DAC complexity. The DUE
 the DAC. Every current source, i.e. one DUE, is proved to be                is related to the number of input bits and the results shown
 directly related to the input bits. If NY is termed as number               in Fig. 5 present a relation between SNR and the number
 of bits contained in Y , reffer to the Fig. 2, then the number              of DAC bits NY . With larger value of NY , higher SNR can
 of output bits from the final adder are                                      be obtained. These results also show a comparison of SNR
                                                                             obtained for different values of NQ and that of the case when
                              NQ + 1    if NQ ≥ NM                           no noise shaping is done.
                 NY =
                              NM + 1    if NQ < NM ,                            The fluctuation in the effective number of bits (ENOB)
                                                                             is observed in terms of SNR. These variations, for different
 where NQ is the word length of XQ . The values of NQ                        values of NQ , are plotted against higher complexity of the
 and NM can be different depending on the quantizer in the                   modulator, as shown in Fig. 6. The ENOB for NQ = 3 shows
 the modulator and XM , respectively. To reduce the DAC                      almost constant value around 14 for the DAC complexity up
 complexity in architecture A1, architecture A2, as given in                 to 10 bits. However, for 9 and 10-bit DAC the ENOB of one-
 Fig. 3, is introduced. In this architecture, the DAC E is                   bit quantizer is same to that of 2-bit. The trade off in terms of
 replaced by B1 and B2 and the digital adder by the analog                   ENOB and DAC’s complexity is that: with higher complexity
 one. The number of DACs is doubled, but the cumulative                      of DAC (or close to k value of NY ) one-bit quantizer is fair
 number of required DUEs is reduced. The required numbers                    choice compare to that of 2-bit.
 of DUE in A1 and A2 architectures are computed as                              Conclusively, for architecture A1, the parameter values
                                                                             corresponding to NL , close to k, seem to be best in terms
                              2NY − 1               if A1                    of trade-offs.
                 DUE =
                              2NQ + 2NM − 2         if A2.                      The complexity reduction in A2 compared to that of A1 in
                                                                             terms of DUE is compiled in Table I. Complexity reduction of
   By mathematical induction, we have that 2NY must always                   Σ∆ with respect to its number of input bits (NL ) are compared
 be greater than 2NQ + 2NM − 1.                                              with the required number of DUE for both the architectures
                                                                             in this table. Table shows number of DUE for three values of
                                                                             NQ . Also note that here one-bit DAC is considered to have 2
                                                                             DUEs.

        XN                         XM                              Output
                                                             +
Input                                          DAC                                                  BW = 40MHz, L = 2, OSR = 50, A1 architecture
                      truncator                                                        120
                                                B1

             +
                  -                                          DAC                       110
                                                              B2
                         XL                    XQ
                                     ∑               a
                                                                                       100
                                                                             SNR[dB]




                                                                                        90
 Fig. 3. To reduce DAC complexity, an architecture termed here as A2 which
 utilizes two DACs with analog addition at the output, is introduced.
                                                                                        80



                      IV. S IMULATION R ESULTS                                          70                                                             NQ = 1
                                                                                                                                                       NQ = 2
    Some behavioral-level simulations results are presented and                                                                                        NQ = 3
 analyzed in this section. Architecture A1 is simulated for                             60
                                                                                          16   14   12        10          8          6        4    2            0
 different word lengths and quantizers, i.e., NM , NL and NQ .                                                     Σ∆ Input bits, NL
 The simulated SNR values, as shown in Fig. 4 present an                     Fig. 4. Comparison between SNR and modulator complexity in terms of
 improvement after the k number of NL . In all of the given                  number of input bits.
 plots, where XN consists of N = 16 bits, this factor k is
TABLE I
                                    Bandwidth = 40MHz, L = 2, OSR = 50, A1 architecture
                                                                                                        C OMPARISON BETWEEN ARCHITECTURES A1 AND A2 WITH RESPECT TO
                                                                                                                               COMPLEXITY.
           110                                                                                                 Required DUE in A1 and     A2 architectures for NQ
                                                                                                                      NQ = 1              NQ = 2               NQ = 3
           100                                                                                                 NL     A1      A2          A1        A2         A1     A2
                                                                                                               16     2       2           4         4          8      8
                                                                                                               15     4       4           8         6          16     10
            90                                                                                                 14     8       6           8         8          16     12
                                                                                                               13     16      10          16        12         16     16
 SNR[dB]




            80                                                                                                 12     32      18          32        20         32     24
                                                                                                               11     64      34          64        36         64     40
                                                                                                               10     128     66          128       68         128    72
            70                                                                                                 9      256     130         256       132        256    136
                                                                                                               8      512     258         512       260        512    264
            60                                                                                                 7      1024    514         1024      516        1024   520
                                                                                                               6      2048    1026        2048      1028       2048   1032
                                                                                     Q=3                       5      4096    2050        4096      2052       4096   2056
            50                                                                       Q=2                       4      8192    4098        8192      4100       8192   4104
                                                                                     Q =1
                                                                                                               3      16384   8194        16384     8196       16384  8200
                                                                                     No noise shaping
            40                                                                                                 2      32768   16386       32768     16388      32768  16392
                  2         4            6          8          10        12         14        16               1      65536   32768       65536     32770      65536  32774
                                                        DAC bits, NY
   Fig. 5. Comparison of SNR with increased DAC complexity and obtained
   resolution provided different quantizers in the A1 architecture.

                                                                                                        use the simulated results to select the most suitable parameter
                                   The fluctuation of ENOB arround the complexity of DAC
            15                                                                                          k to trade off between hardware cost and resolution in the
                          NQ = 1
                                                                                                        system.
           14.5           NQ = 2
                                                                                                           An investigation of the hardware complexity in terms of
                          NQ = 3
            14
                                                                                                        the number of real operations and the impact of possible
                                                                                                        replacement of the analog adder in A2 architecture are left
           13.5                                                                                         for future work.
            13                                                                                                                        R EFERENCES
ENOB




           12.5                                                                                          [1] S. J. Park, R. M. Gray and W. Chou, “Analysis of a sigma delta modu-
                                                                                                             lator with a multi-level quantizer and single-bit feedback,” International
            12
                                                                                                             conference on Acoustics, Speech, and Signal Processing, 1991.
                                                                                                         [2] V. Mladenov, H. Hegt and A. V. Roermund, “On the stability of high
                                                                                                             order Sigma-Delta modulators,” International conference on Electronics,
           11.5
                                                                                                             Circuits and Systems, 2001.
                                                                                                         [3] S. Noli, A. P. Perez, E. Bonizzoni and F. Maloberti, “Delta-Sigma time
            11
                                                                                                             interleaved current steering DAC with dynamic elements matching,”
                                                                                                             52nd IEEE International Midwest Symposium on Circuits and Systems,
           10.5                                                                                              2009.
                                                                                                         [4] S. Brigati, F. Francesconi, V. Liberali and F. Maloberti, P. Malcovati,
                      3         4            5          6         7           8           9        10
                                                            NY                                               M. Poletti, “Design considerations on very high resolution sigma-delta
                          Fig. 6.      ENOB variation with respect to DAC bits.                              modulators,” Proceedings of the 40th Midwest Symposium on Circuits
                                                                                                             and Systems, 1997.
                                                                                                         [5] X. Wang, U. Moon, M. Liu and G. C. Temes, “Digital correlation
                                                                                                             technique for the estimation and correction of DAC errors in multibit
                                                 V. C ONCLUSION                                              mash Σ∆ ADCs,” IEEE International Symposium on Circuits and
                                                                                                             Systems, vol.4, 2002.
      This paper illustrates a valuable trade-off for the design of                                      [6] Wai Lee, “A 4-channel, 18 b Σ∆ modulator IC with chopped-offset sta-
                                                                                                             bilization,” IEEE International Solid-State Circuits Conference, Digest
   Σ∆ modulators where the complexity of digital logic must                                                  of Technical Papers, vol., no., pp.238-239, 453, Feb 1996.
   be kept low to guarantee high-speed operation. The reduction                                          [7] J.W. Bruce and P. Stubberud, “A comparison of hardware efficient
   of modulator complexity for different combinations of input                                               dynamic element matching networks for digital to analog converters,”
                                                                                                             Proceedings of the 43rd IEEE Midwest Symposium on Circuits and
   bits in terms of SNR is presented. The improvement in terms                                               Systems, 2000.
   of SNR is shown to be vital with a slight modification of                                              [8] S. R. Norsworthy, R. Schreier and G. C. Temes, Delta-Sigma Data
   the architecture given in [3]. The A1 architecture is briefly                                              Converters: Theory, Design, and Simulation, IEEE Press, 1996.
                                                                                                         [9] R. Schreier and G. C. Temes, Understanding Delta-Sigma Data Con-
   discussed in terms of DAC complexity and ENOB as well.                                                    verters, IEEE Press, 2005.
   With the maintained number of ENOB DAC complexity is                                                 [10] N. Maghari and G. C. Temes, U. Moon, “Noise-shaped integrating
   successfully reduced to half (considering the number of unit                                              quantisers in Σ∆ modulators,” Electronics Letters , vol.45, no.12,
                                                                                                             pp.612-613, June 4 2009.
   elements required in the DAC) in the proposed, modified                                               [11] J. Candy and G. Temes, Oversampling Delta-Sigma Data Converters,
   architecture A2.                                                                                          New York: IEEE Press, 1991.
      The paper demonstrates that the architectural choice for the
   designer should be the A2 architecture and the designer can

More Related Content

What's hot

First order sigma delta modulator with low-power
First order sigma delta modulator with low-powerFirst order sigma delta modulator with low-power
First order sigma delta modulator with low-powereSAT Publishing House
 
Sigma Delta ADC for Implantable Cardiac Sensing
Sigma Delta ADC for Implantable Cardiac SensingSigma Delta ADC for Implantable Cardiac Sensing
Sigma Delta ADC for Implantable Cardiac SensingRamprasad Vijayagopal
 
A Dual Tree Complex Wavelet Transform Construction and Its Application to Ima...
A Dual Tree Complex Wavelet Transform Construction and Its Application to Ima...A Dual Tree Complex Wavelet Transform Construction and Its Application to Ima...
A Dual Tree Complex Wavelet Transform Construction and Its Application to Ima...CSCJournals
 
Optimization of Packet Length for Two Way Relaying with Energy Harvesting
Optimization of Packet Length for Two Way Relaying with Energy HarvestingOptimization of Packet Length for Two Way Relaying with Energy Harvesting
Optimization of Packet Length for Two Way Relaying with Energy HarvestingIJCNCJournal
 
Cancellation of Zigbee interference in OFDM based WLAN for multipath channel
Cancellation of Zigbee interference in OFDM based WLAN for multipath channelCancellation of Zigbee interference in OFDM based WLAN for multipath channel
Cancellation of Zigbee interference in OFDM based WLAN for multipath channelIDES Editor
 
PPT Image Analysis(IRDE, DRDO)
PPT Image Analysis(IRDE, DRDO)PPT Image Analysis(IRDE, DRDO)
PPT Image Analysis(IRDE, DRDO)Nidhi Gopal
 
Fixed Point Realization of Iterative LR-Aided Soft MIMO Decoding Algorithm
Fixed Point Realization of Iterative LR-Aided Soft MIMO Decoding AlgorithmFixed Point Realization of Iterative LR-Aided Soft MIMO Decoding Algorithm
Fixed Point Realization of Iterative LR-Aided Soft MIMO Decoding AlgorithmCSCJournals
 
Nyquist criterion for distortion less baseband binary channel
Nyquist criterion for distortion less baseband binary channelNyquist criterion for distortion less baseband binary channel
Nyquist criterion for distortion less baseband binary channelPriyangaKR1
 
Multiuser MIMO Vector Perturbation Precoding
Multiuser MIMO Vector Perturbation PrecodingMultiuser MIMO Vector Perturbation Precoding
Multiuser MIMO Vector Perturbation Precodingadeelrazi
 
On Complex Enumeration for Multiuser MIMO Vector Precoding
On Complex Enumeration for Multiuser MIMO Vector PrecodingOn Complex Enumeration for Multiuser MIMO Vector Precoding
On Complex Enumeration for Multiuser MIMO Vector PrecodingTSC University of Mondragon
 
A Simulation Training for Sigma-Delta Modulators by Matlab CAD-Tool
A Simulation Training for Sigma-Delta Modulators by Matlab CAD-ToolA Simulation Training for Sigma-Delta Modulators by Matlab CAD-Tool
A Simulation Training for Sigma-Delta Modulators by Matlab CAD-ToolMCI
 
Thresholding eqns for wavelet
Thresholding eqns for waveletThresholding eqns for wavelet
Thresholding eqns for waveletajayhakkumar
 
Signal and image processing on satellite communication using MATLAB
Signal and image processing on satellite communication using MATLABSignal and image processing on satellite communication using MATLAB
Signal and image processing on satellite communication using MATLABEmbedded Plus Trichy
 
Performance Evaluation of SAR Image Reconstruction on CPUs and GPUs
Performance Evaluation of SAR Image Reconstruction on CPUs and GPUsPerformance Evaluation of SAR Image Reconstruction on CPUs and GPUs
Performance Evaluation of SAR Image Reconstruction on CPUs and GPUsFisnik Kraja
 
On The Fundamental Aspects of Demodulation
On The Fundamental Aspects of DemodulationOn The Fundamental Aspects of Demodulation
On The Fundamental Aspects of DemodulationCSCJournals
 

What's hot (20)

First order sigma delta modulator with low-power
First order sigma delta modulator with low-powerFirst order sigma delta modulator with low-power
First order sigma delta modulator with low-power
 
SigmaDeltaADC
SigmaDeltaADCSigmaDeltaADC
SigmaDeltaADC
 
Iy2415661571
Iy2415661571Iy2415661571
Iy2415661571
 
45
4545
45
 
Sigma Delta ADC for Implantable Cardiac Sensing
Sigma Delta ADC for Implantable Cardiac SensingSigma Delta ADC for Implantable Cardiac Sensing
Sigma Delta ADC for Implantable Cardiac Sensing
 
A Dual Tree Complex Wavelet Transform Construction and Its Application to Ima...
A Dual Tree Complex Wavelet Transform Construction and Its Application to Ima...A Dual Tree Complex Wavelet Transform Construction and Its Application to Ima...
A Dual Tree Complex Wavelet Transform Construction and Its Application to Ima...
 
Optimization of Packet Length for Two Way Relaying with Energy Harvesting
Optimization of Packet Length for Two Way Relaying with Energy HarvestingOptimization of Packet Length for Two Way Relaying with Energy Harvesting
Optimization of Packet Length for Two Way Relaying with Energy Harvesting
 
Gh2411361141
Gh2411361141Gh2411361141
Gh2411361141
 
Cancellation of Zigbee interference in OFDM based WLAN for multipath channel
Cancellation of Zigbee interference in OFDM based WLAN for multipath channelCancellation of Zigbee interference in OFDM based WLAN for multipath channel
Cancellation of Zigbee interference in OFDM based WLAN for multipath channel
 
Gq2411921196
Gq2411921196Gq2411921196
Gq2411921196
 
PPT Image Analysis(IRDE, DRDO)
PPT Image Analysis(IRDE, DRDO)PPT Image Analysis(IRDE, DRDO)
PPT Image Analysis(IRDE, DRDO)
 
Fixed Point Realization of Iterative LR-Aided Soft MIMO Decoding Algorithm
Fixed Point Realization of Iterative LR-Aided Soft MIMO Decoding AlgorithmFixed Point Realization of Iterative LR-Aided Soft MIMO Decoding Algorithm
Fixed Point Realization of Iterative LR-Aided Soft MIMO Decoding Algorithm
 
Nyquist criterion for distortion less baseband binary channel
Nyquist criterion for distortion less baseband binary channelNyquist criterion for distortion less baseband binary channel
Nyquist criterion for distortion less baseband binary channel
 
Multiuser MIMO Vector Perturbation Precoding
Multiuser MIMO Vector Perturbation PrecodingMultiuser MIMO Vector Perturbation Precoding
Multiuser MIMO Vector Perturbation Precoding
 
On Complex Enumeration for Multiuser MIMO Vector Precoding
On Complex Enumeration for Multiuser MIMO Vector PrecodingOn Complex Enumeration for Multiuser MIMO Vector Precoding
On Complex Enumeration for Multiuser MIMO Vector Precoding
 
A Simulation Training for Sigma-Delta Modulators by Matlab CAD-Tool
A Simulation Training for Sigma-Delta Modulators by Matlab CAD-ToolA Simulation Training for Sigma-Delta Modulators by Matlab CAD-Tool
A Simulation Training for Sigma-Delta Modulators by Matlab CAD-Tool
 
Thresholding eqns for wavelet
Thresholding eqns for waveletThresholding eqns for wavelet
Thresholding eqns for wavelet
 
Signal and image processing on satellite communication using MATLAB
Signal and image processing on satellite communication using MATLABSignal and image processing on satellite communication using MATLAB
Signal and image processing on satellite communication using MATLAB
 
Performance Evaluation of SAR Image Reconstruction on CPUs and GPUs
Performance Evaluation of SAR Image Reconstruction on CPUs and GPUsPerformance Evaluation of SAR Image Reconstruction on CPUs and GPUs
Performance Evaluation of SAR Image Reconstruction on CPUs and GPUs
 
On The Fundamental Aspects of Demodulation
On The Fundamental Aspects of DemodulationOn The Fundamental Aspects of Demodulation
On The Fundamental Aspects of Demodulation
 

Similar to 48

Lect2 up390 (100329)
Lect2 up390 (100329)Lect2 up390 (100329)
Lect2 up390 (100329)aicdesign
 
Multiband Transceivers - [Chapter 4] Design Parameters of Wireless Radios
Multiband Transceivers - [Chapter 4] Design Parameters of Wireless RadiosMultiband Transceivers - [Chapter 4] Design Parameters of Wireless Radios
Multiband Transceivers - [Chapter 4] Design Parameters of Wireless RadiosSimen Li
 
First order sigma delta modulator with low-power consumption implemented in a...
First order sigma delta modulator with low-power consumption implemented in a...First order sigma delta modulator with low-power consumption implemented in a...
First order sigma delta modulator with low-power consumption implemented in a...eSAT Journals
 
Implementation performance analysis of cordic
Implementation performance analysis of cordicImplementation performance analysis of cordic
Implementation performance analysis of cordiciaemedu
 
Optical Spatial Modulation with Transmitter-Receiver Alignments
Optical Spatial Modulation with Transmitter-Receiver AlignmentsOptical Spatial Modulation with Transmitter-Receiver Alignments
Optical Spatial Modulation with Transmitter-Receiver AlignmentsMarwan Hammouda
 
129966862758614726[1]
129966862758614726[1]129966862758614726[1]
129966862758614726[1]威華 王
 
PARASITIC-AWARE FULL PHYSICAL CHIP DESIGN OF LNA RFIC AT 2.45GHZ USING IBM 13...
PARASITIC-AWARE FULL PHYSICAL CHIP DESIGN OF LNA RFIC AT 2.45GHZ USING IBM 13...PARASITIC-AWARE FULL PHYSICAL CHIP DESIGN OF LNA RFIC AT 2.45GHZ USING IBM 13...
PARASITIC-AWARE FULL PHYSICAL CHIP DESIGN OF LNA RFIC AT 2.45GHZ USING IBM 13...Ilango Jeyasubramanian
 
Performances des turbo codes parallèles pour un canal satellite non linéaire
Performances des turbo codes parallèles pour un canal satellite non linéairePerformances des turbo codes parallèles pour un canal satellite non linéaire
Performances des turbo codes parallèles pour un canal satellite non linéaireRachidz
 
Design of Low Power Sigma Delta ADC
Design of Low Power Sigma Delta ADCDesign of Low Power Sigma Delta ADC
Design of Low Power Sigma Delta ADCVLSICS Design
 
Cooperative partial transmit sequence for papr reduction in space frequency b...
Cooperative partial transmit sequence for papr reduction in space frequency b...Cooperative partial transmit sequence for papr reduction in space frequency b...
Cooperative partial transmit sequence for papr reduction in space frequency b...IAEME Publication
 
Optimal pilot symbol power allocation in lte
Optimal pilot symbol power allocation in lteOptimal pilot symbol power allocation in lte
Optimal pilot symbol power allocation in lteDat Manh
 

Similar to 48 (20)

Pcm
PcmPcm
Pcm
 
Bh31403408
Bh31403408Bh31403408
Bh31403408
 
Lect2 up390 (100329)
Lect2 up390 (100329)Lect2 up390 (100329)
Lect2 up390 (100329)
 
Multiband Transceivers - [Chapter 4] Design Parameters of Wireless Radios
Multiband Transceivers - [Chapter 4] Design Parameters of Wireless RadiosMultiband Transceivers - [Chapter 4] Design Parameters of Wireless Radios
Multiband Transceivers - [Chapter 4] Design Parameters of Wireless Radios
 
MSc Presentation
MSc PresentationMSc Presentation
MSc Presentation
 
First order sigma delta modulator with low-power consumption implemented in a...
First order sigma delta modulator with low-power consumption implemented in a...First order sigma delta modulator with low-power consumption implemented in a...
First order sigma delta modulator with low-power consumption implemented in a...
 
Ros Gra10
Ros Gra10Ros Gra10
Ros Gra10
 
Implementation performance analysis of cordic
Implementation performance analysis of cordicImplementation performance analysis of cordic
Implementation performance analysis of cordic
 
Optical Spatial Modulation with Transmitter-Receiver Alignments
Optical Spatial Modulation with Transmitter-Receiver AlignmentsOptical Spatial Modulation with Transmitter-Receiver Alignments
Optical Spatial Modulation with Transmitter-Receiver Alignments
 
36
3636
36
 
129966862758614726[1]
129966862758614726[1]129966862758614726[1]
129966862758614726[1]
 
B0160709
B0160709B0160709
B0160709
 
1568973267 effect of multi-tone
1568973267 effect of multi-tone1568973267 effect of multi-tone
1568973267 effect of multi-tone
 
PARASITIC-AWARE FULL PHYSICAL CHIP DESIGN OF LNA RFIC AT 2.45GHZ USING IBM 13...
PARASITIC-AWARE FULL PHYSICAL CHIP DESIGN OF LNA RFIC AT 2.45GHZ USING IBM 13...PARASITIC-AWARE FULL PHYSICAL CHIP DESIGN OF LNA RFIC AT 2.45GHZ USING IBM 13...
PARASITIC-AWARE FULL PHYSICAL CHIP DESIGN OF LNA RFIC AT 2.45GHZ USING IBM 13...
 
Performances des turbo codes parallèles pour un canal satellite non linéaire
Performances des turbo codes parallèles pour un canal satellite non linéairePerformances des turbo codes parallèles pour un canal satellite non linéaire
Performances des turbo codes parallèles pour un canal satellite non linéaire
 
Design of Low Power Sigma Delta ADC
Design of Low Power Sigma Delta ADCDesign of Low Power Sigma Delta ADC
Design of Low Power Sigma Delta ADC
 
B0160709
B0160709B0160709
B0160709
 
Cooperative partial transmit sequence for papr reduction in space frequency b...
Cooperative partial transmit sequence for papr reduction in space frequency b...Cooperative partial transmit sequence for papr reduction in space frequency b...
Cooperative partial transmit sequence for papr reduction in space frequency b...
 
Optimal pilot symbol power allocation in lte
Optimal pilot symbol power allocation in lteOptimal pilot symbol power allocation in lte
Optimal pilot symbol power allocation in lte
 
Hr3114661470
Hr3114661470Hr3114661470
Hr3114661470
 

More from srimoorthi (20)

94
9494
94
 
87
8787
87
 
84
8484
84
 
83
8383
83
 
82
8282
82
 
75
7575
75
 
73
7373
73
 
72
7272
72
 
70
7070
70
 
69
6969
69
 
68
6868
68
 
63
6363
63
 
62
6262
62
 
61
6161
61
 
60
6060
60
 
59
5959
59
 
57
5757
57
 
56
5656
56
 
50
5050
50
 
55
5555
55
 

48

  • 1. Study of Modified Noise-Shaper Architectures for Oversampled Sigma-Delta DACs Nadeem Afzal J Jacob Wikner Department of Electrical Engineering, Department of Electrical Engineering, Link¨ ping University, SE-581 83 Link¨ ping, Sweden o o Link¨ ping University, SE-581 83 Link¨ ping, Sweden o o E-mail: nadeem@isy.liu.se E-mail: Jacob.Wikner@liu.se Abstract—In this paper, modified low-complex, hybrid archi- results are found in Section IV. Finally, we conclude the results tectures for digital, oversampled sigma-delta digital-to-analog and elaborate on some future extensions to the work. converters (Σ∆DACs) are explored in terms of signal-to-noise In this paper, further exploration and modification of the ratio (SNR) and subDAC complexity. The studied techniques illustrate the trade-off in terms of noise-shaper and DAC im- work in [3] is done and some novel results are presented. plementation complexity and loss in SNR. It is found that a fair amount of improvement in SNR is achieved by maintaining low- II. S IGMA D ELTA M ODULATOR complexity of noise shaper. The complexity of the subDAC is yet A simplified interpolation chain for digital-to-analog con- a parameter, directly related to the number of output bits from version including a noise-shaping Σ∆ modulator is shown in the noise shaper. Two different architectures are investigated with respect to Fig. 1. We use a slightly unorthodox starting point to describe subDAC complexity and noise shaper complexity. It is shown the theoretically achievable SNR in the signal path if the the that the required number of DAC unit elements (DUE) can be noise-shaping module is omitted: reduced to half. Index Terms—Sigma-Delta-Modulator, Noise Shaper, DAC- SNR = log2 (DUE), (1) Complexity, Modulator’s Complexity, Hybrid architecture. where DUE is the required number of unit elements in the I. I NTRODUCTION DAC [7]. For the “standard” Nyquist converter, i.e., interpo- lation factor is 1 in the signal chain), DUE would simply be The sigma-delta modulator (Σ∆), among many other mod- given by ulators, attracts remarkable appreciation mainly due to the po- DUE = 2N , (2) tential to reach high accuracy, i.e., high resolution. It achieves high signal-to-noise ratio (SNR) by utilizing a high oversam- and the obtainable (theoretically) SNR at the input of the DAC pling ratio (OSR), where OSR is defined as the ratio between would be the sampling frequency and the baseband signal bandwidth. SNR = 6.02 · N + 1.76 dB. (3) To keep OSR constant for high bandwidths applications, the sampling frequency naturally has to be increased. An increased So, in short: for a Nyquist rate converter, to achieve high sampling frequency is restricted, not only by the mismatches SNR, a high number of bits, N , is required implying a very because of the process technologies but also by the increase complex DAC for high-speed applications. In the case of a in complexity of hardware circuitry. limited bandwidth compared to the Nyquist frequency, i.e., In cases of wideband signals and/or lower OSR applications, oversampling is possible, the SNR can be increased by a factor the SNR can be improved by increasing the number of output 10 · log10 (OSR) [8] implying that bits of the noise shaper, i.e., the number of quantizer bits. SNR = 6.02 · N + 1.67 + 10 · log10 (OSR) dB. (4) This is done at the expense of precision component mismatch in the output digital-to-analog converter (DAC) [1]. One However, just utilizing a high OSR is not always possible can also increase the order of the Σ∆ to further achieve since, still, high-speed architectures are required. Therefore the better performance, but the increment in the order makes the number of bits fed to the DAC is reduced by employing the large domain of parameters difficult to be optimized. Hence, noise shaper. It can effectively be used to maintain the same a higher-order Σ∆ is more difficult to design in terms of SNR within the signal band eventhough the number of bits stability [2]. Conclusively, in the low-OSR scenario, high is lower. The number of bits is reduced by re-quantization at resolution is difficult to reach with a single-bit quantizer - coarser levels while the additional quantization noise is shifted even with the high-order Σ∆ [4]. to frequencies outside the effective signal band. Using an ideal Σ∆ the in-band SNR would be formulated as [9] In Section II we describe some of the mathematical theory (2L + 1)(OSR)2L+1 behind the obtainable Σ∆ modulators. The hybrid noise- SNR = 6.02 · NQ + 10 · log10 + 1.76 dB, π 2L shaper architecture is described in Section III and simulation (5) 978-1-4244-8971-8/10$26.00 c 2010 IEEE
  • 2. part XM from the input XN . In other words, XM consists the of MSBs of XN , whereas, XL consists of the LSBs of Digital Analog XN . The output of the modulator is multiplied by a constant coefficient, a, and then added to the XM which was passed ↑ N NQ through unaffected. The output bits of this adder can now be OSR ∑ DAC fed to the DAC. The second-order Σ∆, in this architecture A1, is a so-called signal feed-back architecture with uniform quantization, as given in [8]. Since the LSBs are processed through the modulator, a right-shift register is employed at its output to match the bit positions of XM and XQ . The relation between the input and the output is given as [3] + + z-1 + + - - z-1 Y = XM + aXQ , (6) where, XM is the signal fed forward at the final adder, a has the scalar value a ≤ 1. XQ is the output of the modulator, Fig. 1. A simplified interpolation chain from digital to analog domain. which in the z domain can be expressed as 2 qn XQ = XL · z −1 + 1 − z −1 · , (7) 2XM where NQ is the number of bits in the coarse quantizer and L is the order of the Σ∆. The gradient of shift in where qn /2XM is the new quantization noise being pushed quantization noise with respect to frequency is dependent upon towards higher frequencies [3]. Now, from the relations given the modulator order such that the higher the order the steeper in (6) and (7), the signal at the output is the slope [10]. Now, we can utilize the L and NQ parameters to achieve an SNR from (5) matching that of (4) where NQ < N . XB = (XM + a · XL ) · z −1 . (8) However, we must also be careful to not increase the com- From (7) it is concluded that coefficient a reduces the am- plexity of the Σ∆ too much. For very high-speed applications plitude of XL and qn /2XM , whereas the XM is unaffected. the power consumption will be very high and obtainable For very small values of XL (compared to those of XM ), maximum speed of the modulator might be limited by word- the quantization noise is further reduced by a while the length, etc. signal power mainly is preserved by XM . This signal power In short, the output number of bits needs to be low, the preservation combined with the reduction of quantization noise order should be low and also the word-length in adders and thus increases SNR. In the case when XM is not sufficiently multipliers must be kept at a minimum. Naturally, there is larger than XL , coefficient a not only reduces the quantization a delicate relation between all these parameters such that noise but also the major part of the signal power. We can a reasonably high SNR can be obtained at reasonably high conclude that coefficient a must be selected according to the hardware and power costs. following relation: III. H YBRID A RCHITECTURE 1 if NL ≥ k a= A hybrid Σ∆ modulator architecture, as presented in for 1/2NM if NL < k, example [3], reduces the number of bits in the path to the actual noise shaper, but still achieves fairly good performance. where, NL and NM are the word lengths of XL and XM , There is no impact on the stability analysis, since the structure respectively. The factor k, is chosen dependent on the strength introduces a feed-forward path rather then feed-back path. of XL and the number of quantizer bits XQ . In this architecture, the number of input bits to the noise shaper is reduced at the cost of some additional in-band quantization noise. In this way the architecture in [3] decreases Input XN XM Y DAC Output the complexity of the modulator, adders and multipliers be- truncator + E come smaller. However, more bits must be allocated to the - mixed-signal DAC at the output. We simply trade the digital + a complexity against the analog complexity. By investigating the XL XQ modified architecture a nice set of trade-off curves is obtained ∑ illustrating these complexity trade-offs, both in terms of SNR and DAC complexity. The modified architecture, denoted in this report as the A1 architecture, is illustrated in Fig. 2. The modulator takes XL Fig. 2. The A1 architecture, as in [3] with reduced complexity of modulator. as input which is obtained by subtracting the LSBs truncated
  • 3. A. Sub DAC empirically selected as  At the output of the interpolation chain, the bits from the 11 if XQ = 1  noise shaper are fed to the DAC. For high-speed applications k = 9 if XQ = 2 the current-steering DAC architecture is chosen since it pro-  7 if XQ = 3.  vide high efficiency and can be implemented for high speed. Typically, the current-steering DACs are composed of unit The simulation results show that for less complex modula- current sources where every current source corresponds to one tors, higher SNR can be achieved. The improvement of course of the 2(∗) levels, where “(∗)” is the number of input bits to comes at some cost, which is the DAC complexity. The DUE the DAC. Every current source, i.e. one DUE, is proved to be is related to the number of input bits and the results shown directly related to the input bits. If NY is termed as number in Fig. 5 present a relation between SNR and the number of bits contained in Y , reffer to the Fig. 2, then the number of DAC bits NY . With larger value of NY , higher SNR can of output bits from the final adder are be obtained. These results also show a comparison of SNR obtained for different values of NQ and that of the case when NQ + 1 if NQ ≥ NM no noise shaping is done. NY = NM + 1 if NQ < NM , The fluctuation in the effective number of bits (ENOB) is observed in terms of SNR. These variations, for different where NQ is the word length of XQ . The values of NQ values of NQ , are plotted against higher complexity of the and NM can be different depending on the quantizer in the modulator, as shown in Fig. 6. The ENOB for NQ = 3 shows the modulator and XM , respectively. To reduce the DAC almost constant value around 14 for the DAC complexity up complexity in architecture A1, architecture A2, as given in to 10 bits. However, for 9 and 10-bit DAC the ENOB of one- Fig. 3, is introduced. In this architecture, the DAC E is bit quantizer is same to that of 2-bit. The trade off in terms of replaced by B1 and B2 and the digital adder by the analog ENOB and DAC’s complexity is that: with higher complexity one. The number of DACs is doubled, but the cumulative of DAC (or close to k value of NY ) one-bit quantizer is fair number of required DUEs is reduced. The required numbers choice compare to that of 2-bit. of DUE in A1 and A2 architectures are computed as Conclusively, for architecture A1, the parameter values corresponding to NL , close to k, seem to be best in terms 2NY − 1 if A1 of trade-offs. DUE = 2NQ + 2NM − 2 if A2. The complexity reduction in A2 compared to that of A1 in terms of DUE is compiled in Table I. Complexity reduction of By mathematical induction, we have that 2NY must always Σ∆ with respect to its number of input bits (NL ) are compared be greater than 2NQ + 2NM − 1. with the required number of DUE for both the architectures in this table. Table shows number of DUE for three values of NQ . Also note that here one-bit DAC is considered to have 2 DUEs. XN XM Output + Input DAC BW = 40MHz, L = 2, OSR = 50, A1 architecture truncator 120 B1 + - DAC 110 B2 XL XQ ∑ a 100 SNR[dB] 90 Fig. 3. To reduce DAC complexity, an architecture termed here as A2 which utilizes two DACs with analog addition at the output, is introduced. 80 IV. S IMULATION R ESULTS 70 NQ = 1 NQ = 2 Some behavioral-level simulations results are presented and NQ = 3 analyzed in this section. Architecture A1 is simulated for 60 16 14 12 10 8 6 4 2 0 different word lengths and quantizers, i.e., NM , NL and NQ . Σ∆ Input bits, NL The simulated SNR values, as shown in Fig. 4 present an Fig. 4. Comparison between SNR and modulator complexity in terms of improvement after the k number of NL . In all of the given number of input bits. plots, where XN consists of N = 16 bits, this factor k is
  • 4. TABLE I Bandwidth = 40MHz, L = 2, OSR = 50, A1 architecture C OMPARISON BETWEEN ARCHITECTURES A1 AND A2 WITH RESPECT TO COMPLEXITY. 110 Required DUE in A1 and A2 architectures for NQ NQ = 1 NQ = 2 NQ = 3 100 NL A1 A2 A1 A2 A1 A2 16 2 2 4 4 8 8 15 4 4 8 6 16 10 90 14 8 6 8 8 16 12 13 16 10 16 12 16 16 SNR[dB] 80 12 32 18 32 20 32 24 11 64 34 64 36 64 40 10 128 66 128 68 128 72 70 9 256 130 256 132 256 136 8 512 258 512 260 512 264 60 7 1024 514 1024 516 1024 520 6 2048 1026 2048 1028 2048 1032 Q=3 5 4096 2050 4096 2052 4096 2056 50 Q=2 4 8192 4098 8192 4100 8192 4104 Q =1 3 16384 8194 16384 8196 16384 8200 No noise shaping 40 2 32768 16386 32768 16388 32768 16392 2 4 6 8 10 12 14 16 1 65536 32768 65536 32770 65536 32774 DAC bits, NY Fig. 5. Comparison of SNR with increased DAC complexity and obtained resolution provided different quantizers in the A1 architecture. use the simulated results to select the most suitable parameter The fluctuation of ENOB arround the complexity of DAC 15 k to trade off between hardware cost and resolution in the NQ = 1 system. 14.5 NQ = 2 An investigation of the hardware complexity in terms of NQ = 3 14 the number of real operations and the impact of possible replacement of the analog adder in A2 architecture are left 13.5 for future work. 13 R EFERENCES ENOB 12.5 [1] S. J. Park, R. M. Gray and W. Chou, “Analysis of a sigma delta modu- lator with a multi-level quantizer and single-bit feedback,” International 12 conference on Acoustics, Speech, and Signal Processing, 1991. [2] V. Mladenov, H. Hegt and A. V. Roermund, “On the stability of high order Sigma-Delta modulators,” International conference on Electronics, 11.5 Circuits and Systems, 2001. [3] S. Noli, A. P. Perez, E. Bonizzoni and F. Maloberti, “Delta-Sigma time 11 interleaved current steering DAC with dynamic elements matching,” 52nd IEEE International Midwest Symposium on Circuits and Systems, 10.5 2009. [4] S. Brigati, F. Francesconi, V. Liberali and F. Maloberti, P. Malcovati, 3 4 5 6 7 8 9 10 NY M. Poletti, “Design considerations on very high resolution sigma-delta Fig. 6. ENOB variation with respect to DAC bits. modulators,” Proceedings of the 40th Midwest Symposium on Circuits and Systems, 1997. [5] X. Wang, U. Moon, M. Liu and G. C. Temes, “Digital correlation technique for the estimation and correction of DAC errors in multibit V. C ONCLUSION mash Σ∆ ADCs,” IEEE International Symposium on Circuits and Systems, vol.4, 2002. This paper illustrates a valuable trade-off for the design of [6] Wai Lee, “A 4-channel, 18 b Σ∆ modulator IC with chopped-offset sta- bilization,” IEEE International Solid-State Circuits Conference, Digest Σ∆ modulators where the complexity of digital logic must of Technical Papers, vol., no., pp.238-239, 453, Feb 1996. be kept low to guarantee high-speed operation. The reduction [7] J.W. Bruce and P. Stubberud, “A comparison of hardware efficient of modulator complexity for different combinations of input dynamic element matching networks for digital to analog converters,” Proceedings of the 43rd IEEE Midwest Symposium on Circuits and bits in terms of SNR is presented. The improvement in terms Systems, 2000. of SNR is shown to be vital with a slight modification of [8] S. R. Norsworthy, R. Schreier and G. C. Temes, Delta-Sigma Data the architecture given in [3]. The A1 architecture is briefly Converters: Theory, Design, and Simulation, IEEE Press, 1996. [9] R. Schreier and G. C. Temes, Understanding Delta-Sigma Data Con- discussed in terms of DAC complexity and ENOB as well. verters, IEEE Press, 2005. With the maintained number of ENOB DAC complexity is [10] N. Maghari and G. C. Temes, U. Moon, “Noise-shaped integrating successfully reduced to half (considering the number of unit quantisers in Σ∆ modulators,” Electronics Letters , vol.45, no.12, pp.612-613, June 4 2009. elements required in the DAC) in the proposed, modified [11] J. Candy and G. Temes, Oversampling Delta-Sigma Data Converters, architecture A2. New York: IEEE Press, 1991. The paper demonstrates that the architectural choice for the designer should be the A2 architecture and the designer can