RADIOENGINEERING, VOL. 27, NO.

2, JUNE 2018 1

Real-Time Generation
of Standard-Compliant DVB-T Signals
Giuseppe BARUFFA, Luca RUGINI, Fabrizio FRESCURA, Paolo BANELLI

Dept. of Engineering, University of Perugia, Via G. Duranti 93, 06125 Perugia, Italy

{giuseppe.baruffa, luca.rugini, fabrizio.frescura, paolo.banelli}@unipg.it

Submitted July 27, 2017 / Accepted December 6, 2017

Abstract. This paper proposes and discusses two software a host PC and an MDS TM-13 IREF DSP board. A com-
implementations of the DVB-T modulator, using C++ and plete DVB-H transmitter and receiver chain that runs in
MATLAB, respectively. All the key features of the DVB-T real time on a multithreaded DSP was described in [8]. In
standard are included. The C++ DVB-T modulator, incor- 2008, [9] proposed a GNU Radio-based DVB-T modulator
porated into the Iris framework developed by Trinity Col- that integrates a hybrid C++/Python code with a universal
lege of Dublin, works in real time on an Intel Core i7 software radio peripheral (USRP) front-end. A C++ re-
2.4 GHz CPU with the Iris testbed. The MATLAB-based ceiver was then proposed in [10] to reduce the computa-
DVB-T modulator is coupled with a receiver implementa- tional cost with respect to [9]. By means of accurate opti-
tion with channel estimation, equalization, soft-output mization strategies and a memory-based acceleration tech-
demapping and channel decoding. The validation step nique [11], real-time reception and decoding of TV signals
demonstrates that the proposed DVB-T software imple- can be achieved. A memory-efficient DVB-T/H receiver
mentations generate standard-compliant DVB-T signals has been proposed in [12], using a DSP-based and acceler-
that are correctly received by commercially available TV ator-assisted SDR architecture together with multithreading
sets and USB dongles. The software code for the Iris-based and a parallelized version of the C language. A software-
C++ modulator, and for the MATLAB-based modulator defined C receiver implementation for ISDB-T is described
and receiver, has been made publicly available under the in [13]: this receiver can operate on a single frequency
GNU license. segment of the ISDB-T signal, thus allowing to receive
a standard definition TV signal in real time. C++ software
for DVB-T signal reception has been proposed in [14],
[15], [16], [17], which demonstrate the real-time feasibility
Keywords for the whole receiving chain. In [18], the most time-con-
suming processing stages (Viterbi decoder and FFT) have
Digital Video Broadcasting-Terrestrial (DVB-T),
been implemented using a graphic processing unit, whereas
OFDM, software-defined radio (SDR), C++,
the remaining stages still run on the CPU and are imple-
MATLAB
mented in C; however, the Reed-Solomon (RS) decoder
has not been considered in the reception chain. In 2014, we
proposed an USRP-based C DVB-T transmitter that works
1. Introduction in real time and makes use of parallelization strategies such
as multithreading and vectorization [19]. The Brazilian
Digital Video Broadcasting-Terrestrial (DVB-T) is version of ISDB-T has been implemented in [20] at the
a popular broadcasting standard used in Europe and transmission side on a regular PC, using GNU radio and
worldwide [1], [2], [3]. Traditionally, DVB–T transmitters C++, achieving real-time processing.
make use of hardware-based architectures, either custom-
ized or reprogrammable. However, software-defined radio MATLAB software for digital video broadcasting
(SDR) solutions are becoming attractive in order to incor- systems has been proposed as well. For instance, [21] de-
porate many software-defined communication standards velops DVB-T transmission and reception chains, with
into a single device, which can be driven by a low-cost a subsequent SIMD-oriented pre-compilation step that
personal computer (PC) with multicore central processing provides optimized assembly code for specialized repro-
units (CPUs) [4], [5], [6]. grammable processors. Some portions of the DVB-T re-
ception chain have been implemented in MATLAB in [22]
In the last fifteen years, SDR solutions for DVB-T for the purpose of localization of GNSS systems: due to the
systems have been investigated by many researchers, using low complexity of the synchronization algorithm, real-time
different software languages, mainly C++ and MATLAB. processing was achieved. A MATLAB-based implementa-
For what concerns the C++ language, [7] proposed a mixed tion of DVB-T transmitter and receiver is proposed in [23]
SDR digital signal processor (DSP) platform that uses and simulated in [24], focusing on channel coding and

DOI: 10.13164/re.2018.0001 APPLICATIONS OF WIRELESS COMMUNICATIONS

2 G. BARUFFA, L. RUGINI, F. FRESCURA, ET AL., REAL-TIME GENERATION OF STANDARD-COMPLIANT DVB-T SIGNALS

decoding operations. A MATLAB DVB-T receiver is de- Transport Stream (TS) packets, each with 188 data bytes
scribed in [19], including mode detection and time and containing compressed video and audio. Groups of eight
frequency synchronization. TS packets are scrambled using a pseudo-random binary
sequence that produces a scrambled sequence with 1503
This paper presents and describes a new software im-
bytes. The obtained sequence is then parsed and encoded
plementation of the DVB-T modulator using C++. Differ-
using a nonbinary shortened RS code with 188 input bytes
ently from previous literature (e.g., [9] and [19]), the de-
and 204 output bytes (the additional 51 zero bytes are not
signed C++ code is integrated into the Iris SDR testbed
transmitted) [1]. These output bytes are convolutionally
developed by Trinity College of Dublin (TCD) [25]. Actu-
interleaved using I = 12 delay paths and memory cells with
ally, the Iris testbed and the associated software enclose
MI = 204/I = 17 bytes. Then, the interleaved bits are
several features (e.g., runtime reconfiguration, network
convolutionally encoded using a rate rc = 1/2 and a con-
stack support, embedded systems support, component-
straint length L = 7. The code rate can be increased up to
based architecture [25]) and also provide implicit multi-
rc  {2/3, 3/4, 5/6, 7/8} by puncturing (nonequispaced bit
threading capabilities. This way, differently from [19],
decimation) [1]. Successively, bit interleaving and  -bits
real-time transmission can be achieved without applying
specific parallelization strategies to the designed DVB-T symbol interleaving is performed using a RAM whose size
C++ code. For the purpose of validation of the designed depends on the DVB-T mode (6048 cells for 8K, 1512
DVB-T C++ software, we also include the implementation cells for 2K). Afterward, the mapper uses Gray encoding
of a MATLAB-based DVB-T modulator: the validation and square quadrature-amplitude modulation (QAM) with
procedure compares the output of the C++ DVB-T modu- M = 2. The obtained QAM data cells are inserted into
lator with the output of the MATLAB DVB-T modulator. a frame adapter that includes signaling information, such as
The correctness of the validation has been further con- boosted pilots for channel estimation and transmission
firmed by the correct reception, on regular TV sets and parameter signaling (TPS). The whole frame is then con-
USB dongles, of the DVB-T signal transmitted on-air in verted in orthogonal frequency division multiplexing
a laboratory environment. (OFDM) symbols, with cyclic prefix (CP): one frame con-
tains NF = 68 OFDM symbols, and one superframe contains
In addition to a standard-compliant DVB-T modula- four frames. The OFDM modulator is detailed in Sec. 2.1.
tor, the designed MATLAB code also includes a DVB-T If necessary, the DVB-T modulator can include an inter-
receiver with channel estimation, equalization, soft-output polator that adjusts the sample rate and a spectrum emis-
demapping, Viterbi decoding, deinterleaving, Reed-Solo- sion mask (SEM) filter. Interpolation and SEM filtering is
mon (RS) decoding, and descrambling. Differently from described in Sec. 2.2.
most of the previous SDR-based DVB-T systems proposed
in the literature so far, an important feature of the designed
C++ and MATLAB code is the public availability under 2.1 OFDM
the GNU license (at https://github.com/wishful-
Each OFDM symbol contains K active carriers and
project/module_iris/tree/master/dvb-tx-iris), to enable reuse
NFFT – K virtual carriers as guard frequency bands. NFFT is
by other researchers and practitioners.
the FFT size. OFDM performs an inverse FFT that con-
The remainder of this paper is organized as follows. verts the assembled data from the frequency domain to the
Section 2 presents a brief overview of the DVB-T standard. time domain. A CP (with NG samples) is also prepended to
Section 3 describes the implementation and the validation each OFDM symbol. The OFDM baseband signal s[n] can
of the C++ DVB-T modulator, while Section 4 illustrates be expressed by
the MATLAB implementation of DVB-T. Section 5 dis-
 N F 1  N F 1 K 1
cusses the impact of the proposed software, while Sec-
s[n ]    zm ,l [n ]     cm ,l ,k Gk m ,l ,k [n ] (1)
tion 6 concludes the paper. m0 l 0 m0 l 0 k 0

where m is the frame index, l is the OFDM symbol index, k

is the subcarrier index, cm,l,k is the QAM data cell, Gk is
2. Overview of the DVB-T Standard a weighting factor that pre-compensates linear distortions
The DVB-T modulator can be summarized as in caused by following stages (Gk = 1 if no distortion), and
Fig. 1 [1]. The data to be transmitted are MPEG-2 zm,l[n] is the OFDM symbol in the time domain.

Input data Reed-Solomon Convolutional Convolutional Bit

Scrambler Puncturer
(bytes) encoder interleaver encoder interleaver

DVB-T signal Spectrum emission OFDM Frame Symbol

Interpolator Mapper
(samples) mask filter modulator adapter interleaver

Fig. 1. Block diagram of the DVB-T modulator.

RADIOENGINEERING, VOL. 27, NO. 2, JUNE 2018 3

The orthogonal carriers  m ,l , k [n] are expressed by adjacent components by the framework, thus allowing data
to be exchanged between the components. Metadata are
 k  K2  passed among components to specify particular infor-
 m ,l ,k [n ]  exp  j2 n  N G  (l  mN F ) N S  mation data, such as a time stamp or a data rate. The Iris
 N FFT  (2)
 NS n  (l  mN F ) N S  framework has also the task to perform data safety, that is,
one component cannot use data while another component
where K2 = (K – 1)/2, NS = NFFT + NG, and N [n] is equal to is still producing them, and vice-versa. A useful feature of
S

one for n = 0, 1,…, NS – 1 and to zero elsewhere. There are the Iris framework is its capability of parallelization of
3 possible modes for OFDM: NFFT  {2048, 4096, 8192} sequential tasks: as explained in Sec. 3.1, parallelization
for 2K, 4K (in DVB-H only), and 8K, respectively, which can be achieved by assigning group of components (whose
correspond to K  {1705, 3409, 6817} active carriers. The tasks can proceed sequentially) to an engine, i.e., an Iris
CP ratio NG/NFFT is selected in the set {1/32, 1/16, 1/8, 1/4}. framework environment that has its own data buffers and
can run in a thread separate from that of the other engines.
During a typical call to the Iris framework, the host
2.2 Interpolation and SEM Filter application reads a configuration file and assembles a radio
The interpolator is not part of the DVB-T standard [1] structure through the Iris application programmer interface
and is required only when the digital-to-analog converter (API). A component manager is accessed to select the
(DAC) sample rate fs,DAC is different than the DVB-T sam- requested components from a library of precompiled com-
pling rate fs,DVB-T = 64/7 MHz. When fs,DAC > fs,DVB-T, the ponents that include the designed DVB-T software. All the
oversampling ratio is Ro = fs,DAC / fs,DVB-T = Lo/Mo, where Lo selected components are then instantiated and executed by
and Mo are coprime. The sampling rate conversion can be the radio engine, to perform the designed task.
done by an interpolating filter with impulse response hI(t),
producing as output [19]
3.2 Design of the C++ Code Structure
P2
y [n ]   s[ pn  i ]hI  (i  n )Ts,DVB-T  (3) Each DVB-T basic building block has been assigned
i  P1 to a single serial task that runs inside a unique Iris compo-
where pn = nTs,DAC / Ts,DVB-T = n/Ro , n = n/Ro – pn, nent. This way, parallelization can be achieved by instanti-
Ts,DVB-T = 1/fs,DVB-T, and Ts,DAC = 1/fs,DAC . The interpolator ating different components to separate engines. To achieve
distortion can be pre-compensated by choosing the highest degree of parallelization, we choose to fit every
Gk = 1/HI((k – K2) fs,DVB-T/NFFT) in (1), where HI(f), is the component inside its own engine: indeed, this design
Fourier transform of hI(t). philosophy permits a pipeline-like execution of the instruc-
tions of consecutive engines, thereby minimizing the pro-
The SEM filter (not in the DVB-T standard [1]) is re- cessing latencies. Hence, we discarded the alternative
quired only when the SEM has to be imposed at the base- philosophy that fits all components inside a single engine:
band level. This filter also reduces the IF images caused by indeed, this way the DVB-T data flow would be executed
interpolation [19]. The number of taps depends on the serially and would introduce large latencies in the pro-
required stopband attenuation and transition bandwidth, cessing of consecutive data blocks, because the engine
and should be low for real-time operation. would have to wait for the end of the processing of the last
component before reactivating the first component. To
maximize the number of engines, we split the DVB-T
3. C++ DVB-T Modulator modulator in 12 functional components, i.e., the 12 blocks
of Fig. 1. Next, we separately designed each block, as
The implemented C++ DVB-T modulator is struc- explained in Sec. 3.3. As verified in Sec. 3.4, this design
tured with a modular design that makes the designed soft- method enables real-time operation of the DVB-T modu-
ware easily embeddable into existing SDR frameworks. lator. Clearly, this highly parallel design increases the us-
age of computing resources, such as RAM and threads,
with respect to a slower solution that uses a unique engine
3.1 Overview of the Iris Framework shared by all the components.
The Iris SDR framework [25], developed by the Tele-
communications Research Centre at TCD, is primarily
3.3 Design of the C++ Blocks
employed to perform cognitive-radio experiments and
enable dynamic spectrum access [26], so it is highly suita- Every block is designed according to the Iris API in
ble for real-time signal processing operations. The Iris C++ language. Each block is instantiated as a single com-
framework, written in C++, is intended to run on general- ponent in a derived base class of the Iris API specifications
purpose processors and can be configured to perform many [25], thus inheriting all the methods needed for the proper
different radio structures. A radio structure is built on basic management of that component. Component data pro-
blocks called components, which perform a particular pro- cessing methods are implemented as specified by the Iris
cess of the radio chain. A data buffer is positioned between API [25]. Thus, a generic component requests a data buffer
4 G. BARUFFA, L. RUGINI, F. FRESCURA, ET AL., REAL-TIME GENERATION OF STANDARD-COMPLIANT DVB-T SIGNALS

to process, waiting in idle mode until input data are pro-

vided. The designed blocks are generally unable to request
a particular number of input samples or data. All the
blocks, except the scrambler and the SEM filter, have their
own private internal data buffer, where multiple data input
bursts can be stored, until a sufficient number of samples is
collected in order to trigger the designed signal processing
operation.
The C++ code of the first 10 blocks has been
designed by translating the DVB-T standard specifications
[1] into lines of code, avoiding the use of existing libraries:
this way, we have maximum control of the specific opera-
tion of each block. The only exception to this general rule
is the use of the FFTW library [27] inside the OFDM
block, for maximum performance and portability. The
interpolator and the SEM filter have been designed [19]
Fig. 2. PSD of the generated DVB-T signal.
assuming an Ettus USRP N210, whose sample frequency is
fs,DAC = 100/nDAC MHz (nDAC is a positive integer). The Q = 2/2NDAC  1.210–4. Thus, for validation purposes,
designed SEM filter uses a Kaiser window with passband two floating point values are considered equal when their
bandwidth W = (K + 1) fs,DVB-T /NFFT  7.6 MHz, transition difference is below a tolerance value , where  is chosen
bandwidths of 700 kHz, and stopband attenuation of such as  < Q. In this case, the two floating point values
25 dB. will produce samples that either are equal or differ at
maximum by one quantization level Q only. In our tests,
3.4 Validation of the C++ Code  = 10–4.
The code has been validated using the automated Note that the last two optional blocks of the DVB-T
testing provided within the Iris framework building step. modulator of Fig. 1 have not been implemented in
The command cmake with ctest [28] automatically per- MATLAB. Concerning the SEM filter output, Figure 2
forms several tests on both code sanity and expected out- shows the power spectrum density (PSD) of the generated
puts. Concerning the code sanity tests, we performed the DVB-T signal, at a carrier frequency fc = 666 MHz, using
following validation actions: (a) smoke tests to verify that two Ettus USRP N210 (one as transmitter and another one
the code compiles and links correctly into an executable; as spectrum analyzer). Note the flat PSD in the passband
(b) crash-proof tests to verify that the produced executable and the total attenuation of 35 dB in the stopband. The
runs without crashes; (c) black-box tests to verify that the spike at 671 MHz is generated by the local oscillator car-
number and data types of inputs and outputs match the rier, using a frequency offset for near-zero IF configuration
expected ones, without knowledge of the implementation (fIF = 5 MHz): this spike can be filtered out with an analog
of the block; (d) white-box tests to verify that the number filter during the final amplification stage. If a final analog
and data types of inputs and outputs match the expected filter is absent (such as in USRP devices), this spike can be
ones, with knowledge of the implementation of the block. moved into the transition band to avoid adjacent channel
interference (e.g., by using fIF = 3.9 MHz).
Concerning the tests on the expected outputs of the
DVB-T modulator, we have compared the output data of Summarizing, all the different tests performed for
the C++ modulator to a reference output, for a predefined every block of the DVB-T modulator chain have been
set of input data. To obtain the reference outputs, we have completely successful, including all the code sanity tests
implemented the first 10 blocks of Fig. 1 in a high-level and all the expected output tests. In addition, we have also
interpreted language, i.e., MATLAB. For those blocks verified that the generated DVB-T signal, when transmitted
operating on fixed point data (i.e., the first seven blocks of in real time by an Ettus USRP (such as N210, B205, B210)
Fig. 1), we required that the output data samples of the two in a laboratory environment, is correctly received by both
different implementations (C++ design and MATLAB commercial TV sets and USB dongles.
reference) are bit-by-bit equal. Differently, for those blocks
operating on floating-point data (i.e., mapper, framer, and
OFDM), we required that the output data samples of the
3.5 Time Performance
two implementations have a maximum difference below To enable real-time operation of the designed DVB-T
a preselected tolerance. This tolerance depends on the modulator, every block should be able to achieve a pro-
expected quantization step Q, which in turn depends on cessing speed higher than the speed (specified in [1]) re-
the characteristics of the adopted DAC. For instance, quired to perform the block task. Iris contains dedicated
assuming that the real and the imaginary parts of the signal steps for the benchmarking of single processing compo-
are in the range [1, 1] , a DAC with NDAC = 14 bits gives nents, which in our case coincide with the DVB-T blocks.
RADIOENGINEERING, VOL. 27, NO. 2, JUNE 2018 5

This benchmarking suite provides the speed of the block, Configuration SEM filter Ract Rst,max Fsu
to be compared with the time requirement. For every block, 8K, 1/4, 4QAM, 1/2 No 12.85 4.98 2.6
the maximum input data rate Rst,max is constrained by [1] 8K, 1/4, 4QAM, 1/2 23 taps 9.38 4.98 1.9
and by the input sample rate of the SDR device. For the 8K, 1/4, 4QAM, 2/3 No 17.17 6.64 2.6
USRP, we used fs,DAC = 10 MHz (fs,DAC = 12.5 MHz), which 8K, 1/8, 4QAM, 5/6 No 23.63 9.22 2.6
leads to Ro = 35/32 (Ro = 175/128) and to a SEM filter with 8K, 1/32, 16QAM, 1/2 No 30.75 12.06 2.5
19 taps (23 taps). By comparing Rst,max with the actual data 8K, 1/16, 16QAM, 2/3 No 39.48 15.61 2.5
rate Ract measured by the suite, we obtain the speed-up 8K, 1/16, 16QAM, 5/6 No 49.37 19.52 2.5
factor Fsu = Ract / Rst,max of each block. We assume that 8K, 1/4, 64QAM, 3/4 No 57.15 22.39 2.6
8K, 1/4, 64QAM, 3/4 23 taps 40.72 22.39 1.8
real-time operation is possible when, for all the blocks,
8K, 1/8, 64QAM, 5/6 No 70.10 27.65 2.5
Fsu > 1.2, where the 20% margin accounts for additional
8K, 1/32, 64QAM, 7/8 No 80.03 31.67 2.5
latencies due to the data exchange among blocks (operated
8K, 1/32, 64QAM, 7/8 23 taps 57.58 31.67 1.8
by the engines) and overall thread execution scheduling.
2K, 1/32, 64QAM, 7/8 No 80.32 31.67 2.5
Table 1 reports the achieved performance Ract, the re-
Tab. 2. Input data rates and speed-up factor for the whole
quired maximum data rate Rst,max, and the speed-up factor DVB-T modulator. Data rates are expressed in Mb/s.
Fsu of all the blocks. The benchmarks are run on an Intel
Core i7-3630QM CPU at 2.4 GHz with 8 GB DDR3 RAM sider the different DVB-T parameters (number of carriers,
and Ubuntu 16.04 OS, and the programs have been com- CP length, modulation, and code rate) and the absence or
piled with the GNU C++ compiler v. 5.4.0. Benchmarking presence of the optional blocks in Fig. 1, i.e., the interpo-
is performed adopting input data of large size, thus mini- lator and the SEM filter. The output of the last block in the
mizing the impact of the Iris engine overhead on data chain (either the OFDM modulator or the SEM filter) has
buffer management. All the speed-up factors are above the been forwarded to a virtual dummy load (the /dev/null
threshold value 1.2: in Tab. 1, the fastest blocks are the device of the Ubuntu file system). In all cases, we obtained
scrambler and the convolutional interleaver, while the a typical speed-up factor of Fsu = 2.5 in the absence of
slowest blocks are the SEM filter and the RS encoder. interpolation and SEM filtering, and of Fsu = 1.8 in the
presence of interpolation and SEM filtering. Since the
Table 2 reports the achieved time performance for
obtained speed-up factors are greater than 1.2 in all cases,
the whole chain. The different configurations of Tab. 2 con-
we assume that our C++ DVB-T software can work in real
Block Ract Rst,max Fsu time without problems. This is confirmed by all the real-
Scrambler 4158.5 31.7 131.3 time experiments we have performed. Note that the speed-
RS encoder 211.8 31.7 6.7 up factors of Tab. 2 for the aggregate system are signifi-
Convolutional interleaver 4016.4 34.4 116.8 cantly lower than the speed-up factors of the slowest
Convolutional encoder 409.0 34.4 11.9 blocks in Tab. 1. This means that the latencies due to the
Puncturer 1/2 552.0 68.7 8.0 data exchange among blocks and the thread execution
2/3 612.7 68.7 8.9 scheduling times cannot be neglected.
3/4 627.0 68.7 9.1
5/6 616.6 68.7 9.0 When the software runs in real time, typically the
7/8 604.7 68.7 8.8 overall CPU load is about 16%. Therefore, the remaining
Bit interleaver 4QAM 357.3 39.3 9.1 computational resources are sufficient for performing other
16QAM 374.5 39.3 9.5 tasks, such as retrieving or building the transport stream
64QAM 379.1 39.3 9.6 from a local disk, or receiving the transport stream by
Symbol interleaver 2K 317.3 6.5 48.8 a remote network connection. The used memory typically
4K 313.0 6.5 48.2 amounts to roughly 100 MB and is almost constant for all
8K 314.7 6.5 48.4 the configurations. Such computational loads could allow
Mapper 4QAM 485.1 6.5 74.6 a single modern CPU to run several instances of the DVB-
16QAM 480.0 6.5 73.8 T modulator in parallel, each one allocated to a different
64QAM 449.1 6.5 69.1 TV channel, with the only bottleneck consisting in the
Frame Adapter 2K 197.5 6.5 30.4 limited bandwidth of the communication channel used to
4K 191.2 6.5 29.4 convey the generated waveform samples towards the SDR
8K 178.4 6.5 27.4 devices.
OFDM 2K 1/32 122.0 7.4 16.5
8K 1/4 96.4 6.1 15.8
Interpolator 35/32 120.8 9.1 13.2
175/128 103.2 9.1 11.3
4. MATLAB DVB-T Software
SEM filter 19 taps 52.6 10.0 5.3
23 taps 45.7 12.5 3.7
4.1 Modulator
Tab. 1. Input data rates and speed-up factor for the compiled
C++ blocks. Data rates are in Mb/s for the first six As explained in Sec. 3.4, the first 10 blocks of Fig. 1
blocks and in Msample/s for the last six blocks. have been implemented also in MATLAB, for the purpose
6 G. BARUFFA, L. RUGINI, F. FRESCURA, ET AL., REAL-TIME GENERATION OF STANDARD-COMPLIANT DVB-T SIGNALS

of validation of the designed C++ blocks. These MATLAB be computed using different techniques: in our case, we
scripts have been designed independently from the C++ have adopted a Log-max approximation [29], [30], to
blocks, starting from the DVB-T standard [1]. These achieve a good trade-off between decoding accuracy and
scripts have been further validated by comparing their computational complexity. For the generic m-th bit carried
outputs with other MATLAB implementations, such as by the k-th QAM symbol, the Log-max LLR Λ(bk,m) is
[23]. This comparison has revealed a minor typo in the RS expressed as
encoder of [23] (the code generator polynomial in [1] is
different from the MATLAB default one used in [23]). | Cˆ [k ] |2  min Y [k ]   2  min Y [k ]   2  (7)
 (bk , m )    Sm0 
4  Sm1

4.2 Demodulation and Decoding where Sm0 is the subset of constellation points where the
The block diagram of a DVB-T receiver is shown in m-th bit is 0, and Sm1 is the subset of constellation points
Fig. 3. We assume that a previous subsystem has already where the m-th bit is 1.
detected the DVB-T operating parameters (OFDM mode, The outputs of the LLR block are then forwarded to
CP, and code rate) and has properly synchronized the re- the symbol and bit deinterleavers, and then to the depunc-
ceived OFDM symbols, both in time and frequency. For turer. Depuncturing assigns a null value to the LLR of
instance, parameter detection and synchronization could be erased bits, such that the subsequent Viterbi decoder will
performed as in [19]. Thus, the MATLAB receiver re- not consider the contribution of erased bits in the computa-
moves the CP from the received signal, and subsequently tion of branch and path metrics. The Viterbi decoder uses
performs the FFT. After the FFT processing, the noisy soft inputs with block truncation, and should use a trace-
received signal R[k] on the k-th carrier can be expressed by back length lTB several times larger than the code constraint
R[k ]  C[k ]S[k ]  W [k ] (4) length, in order to take into account the effects of punc-
tured bits. The chosen Viterbi algorithm has a hard-decided
where C[k] is the channel transfer function, S[k] is the data output, thus the decoded bits are sent to the convolutional
(or pilot, or TPS) signal, and W[k] is the additive white deinterleaver and to the RS decoder. The RS decoder pads
Gaussian noise (AWGN), on the k-th carrier, and k = with zeroes the input words and follows the usual decoding
0, …, K – 1 represents the active carrier index. Then, the procedure (syndrome computation, error location, error
data carrier samples are separated from the pilot carrier evaluation, and error correction). The decoded message
samples: the information conveyed by the pilot carriers is words are then descrambled and the original TS data are
instrumental to estimate the channel transfer function C[k] recovered.
on the k-th carrier, for equalization purposes. The channel
estimator performs a least-squares interpolation based on
the FFT, as in [2]. In practice, the estimated channel Ĉ[k] is 4.3 Performance Results
a noisy copy of the real channel, as expressed by Figure 4 shows the results obtained by simulations on
Cˆ [k ]  C[k ]  WC [k ] (5) the AWGN and P1 channel [1], represented in terms of bit
error rate (BER) after soft Viterbi decoding versus the
where WC[k] is the channel estimation error, mainly due to carrier-to-noise ratio (C/N). The C/N is defined as
receiver AWGN. The data carriers are then equalized using K 1
the zero-forcing criterion, as
C PR  E{| C[k ]S[k ] | } 2

R[k ]   k 0
(8)
N  W2 K 1
Y [k ]  (6) E{| W [k ] |2 } E{| C[k ] |2 }
Cˆ [k ] k 0

where Y[k] is the noise-affected equalized constellation where PR   k 0 E{| C[k ]S [k ] |2 } 

K 1 K 1
k 0
E{| C [k ] |2 } is the
point and R[k] is the received constellation point expressed
by (4). The equalized data form the basis for the estimation average power of the received signal, σW2 = E{W[k]2} is the
of the log-likelihood ratios (LLRs), which enable the soft- AWGN variance, and E{} is the expectation operator.
input decoding of the convolutional code. These LLRs can The P1 multipath channel (static, non-line-of-sight) leads to

DVB-T signal CP removal Pilot-data Channel LLR Symbol

Equalization
(samples) and FFT separation estimation computation deinterleaver

Output data Reed-Solomon Convolutional Viterbi Bit

Descrambler Depuncturer
(bytes) decoder deinterleaver decoder deinterleaver

Fig. 3. Block diagram of the MATLAB DVB-T receiver.

RADIOENGINEERING, VOL. 27, NO. 2, JUNE 2018 7

-1
10 M CR C/N (AWGN) C/N (P1)
4QAM 1/2
4QAM 2/3 [1] Sim. [1] Sim.
4QAM 3/4
4QAM 5/6
4QAM 7/8
4 1/2 3.5 3.6 5.9 5.9
10
-2
16QAM 1/2
16QAM 2/3
2/3 5.3 5.3 9.6 9.2
16QAM 3/4
16QAM 5/6 3/4 6.3 6.2 12.4 11.6
16QAM 7/8
5/6 7.3 7.2 15.6 14.6
Bit Error Rate (BER)

64QAM 1/2
64QAM 2/3
64QAM 3/4 7/8 7.9 7.9 17.5 15.9
-3 64QAM 5/6
10
64QAM 7/8 16 1/2 9.3 9.3 11.8 11.7
2/3 11.4 11.4 15.3 15.0
3/4 12.6 12.6 18.1 17.3
10
-4
5/6 13.8 13.8 21.3 20.1
7/8 14.4 14.4 23.6 22.1
64 1/2 13.8 13.8 16.4 16.2
-5
2/3 16.7 16.7 20.3 20.0
10
0 5 10 15 20 25 3/4 18.2 18.2 23.0 22.4
C/N (dB)
5/6 19.4 19.4 26.2 25.2
(a)
7/8 20.2 20.2 28.6 27.2
-1
10
4QAM 1/2 Tab. 3. Simulated QEF C/N (in dB) of the DVB-T standard.
4QAM 2/3
4QAM 3/4
4QAM 5/6
AWGN F1 channel P1 channel
4QAM 7/8
M CR
-2
10
16QAM 1/2
16QAM 2/3
QEF BER QEF BER QEF BER
4 1/2 3.5 2.010-4 4.1 1.510-4 5.9 2.010-4
16QAM 3/4
16QAM 5/6
Bit Error Rate (BER)

16QAM 7/8
64QAM 1/2 2/3 5.3 2.010-4 6.1 1.410-4 9.6 1.010-4
64QAM 2/3

-3
64QAM 3/4 3/4 6.3 1.910-4 7.2 1.910-4 12.4 7.310-5
10 64QAM 5/6
64QAM 7/8 5/6 7.3 1.910-4 8.5 1.410-4 15.6 7.810-5
7/8 7.9 1.910-4 9.2 2.310-4 17.5 3.610-5
16 1/2 9.3 1.810-4 9.8 1.710-4 11.8 1.510-4
-4
10 2/3 11.4 2.010-4 12.1 1.610-4 15.3 1.310-4
3/4 12.6 1.910-4 13.4 2.010-4 18.1 7.110-5
5/6 13.8 2.010-4 14.8 1.910-4 21.3 5.210-5
-5 7/8 14.4 2.110-4 15.7 1.610-4 23.6 3.610-5
10
5 10 15 20 25 30 64 1/2 13.8 1.910-4 14.3 1.710-4 16.4 1.410-4
C/N (dB)
(b)
2/3 16.7 2.010-4 17.3 1.910-4 20.3 1.310-4
3/4 18.2 1.910-4 18.9 2.010-4 23.0 1.110-4
Fig. 4. Simulated performance of DVB-T using the MATLAB 5/6 19.4 2.110-4 20.4 1.910-4 26.2 5.810-5
code: (a) AWGN channel; (b) P1 channel.
7/8 20.2 2.110-4 21.3 2.410-4 28.6 3.410-5
Rayleigh fading. We assume perfect channel knowledge, Tab. 4. Simulated BER of DVB-T at the QEF C/N.
and soft Viterbi decoding with unquantized inputs and
a trace-back length of lTB = 128.
According to [1], the performance should be meas- 5. Discussion
ured at the quasi error-free (QEF) condition (not more than
We discuss the impact of the proposed SDR DVB-T
a single uncorrected error per hour). QEF corresponds to
modulator, and we compare the proposed DVB-T software
a BER after the RS decoder of 110–11, and to a BER after
with DVB-T software designs available in the literature.
the Viterbi decoder of 210–4. Table 3 reports the values of
C/N (in dB) for QEF, for different modulation sizes M
and coding rates (CRs). The C/N values obtained using the 5.1 Impact
MATLAB modulator and receiver are in good agreement
with the C/N values provided by the DVB-T specifications First, we explain the advantages of SDR-based
[1]. Note that perfect agreement is virtually impossible, DVB-T transmitters with respect to conventional DVB-T
since the DVB-T specifications do not disclose all the transmitters, and successively we list and discuss some
decoding parameters, such as the trace-back length. Ta- useful applications of the proposed DVB-T software.
ble 4 shows the BER after the Viterbi decoder, simulated at In the upcoming future, some TV channels located in
the QEF C/N value, for different DVB-T configurations the upper UHF band will be switched off, since that spec-
and for different channels, such as AWGN, F1, and P1 [1]. trum will be reallocated to cellular networks for extending
The F1 multipath channel (static, line-of-sight) leads to their 4G (or upcoming 5G) services [31], [32]. This spec-
Rice fading. Table 4 explains that the simulated BER is trum reallocation pushes the broadcasting players (opera-
either close to the QEF BER of 210–4 or lower, thus con- tors and equipment companies) to move their technologies
firming that the decoding parameters have been selected towards SDR-based ones. The reason for this change is
properly. twofold. First, the profitability of hardware-based broad-
8 G. BARUFFA, L. RUGINI, F. FRESCURA, ET AL., REAL-TIME GENERATION OF STANDARD-COMPLIANT DVB-T SIGNALS

casting transmitters will be reduced, because there will be spect to the C++ DVB-T in [19], there are two main differ-
less market opportunities to catch. However, SDR trans- ences. First, the C++ code of [19] is not publicly available.
mitters, due to their easier re-programmability and lower Second, the C++ code of [19] uses vectorization and mul-
development costs (at least in their digital segment), will tithreading, while the designed C++ code uses the intrinsic
allow to accommodate the market uncertainties. Second, parallelization provided by the Iris framework. Therefore,
broadcasting operators could use many low-power small- although both transmitters work in real time, our non-par-
coverage transmitters co-located with cellular network base allelized C++ code is simpler and more easily understanda-
stations, instead of using few high-power large-coverage ble than [19]: this facilitates code modification and code
transmitters. In this case, SDR systems can be shared be- reuse by other researchers.
tween broadcasting and cellular technologies. With respect to the MATLAB DVB-T of [19], there
Our software can be used for different applications are three main differences. First, the designed MATLAB
related to DVB-T, such as signal analysis, interference code is publicly available (at https://github.com/wishful-
estimation, power resource allocation, coverage issues, and project/module_iris/tree/master/dvb-tx-iris/scripts/dvbt/
so on. For instance, the designed software permits a statis- MATLAB). Second, the designed code also includes
tical evaluation of the peak-to-average power ratio of true a validated DVB-T modulator and (F1 and P1) channel
DVB-T signal waveforms. As a second example, adjacent generation functions. Third, although [19] includes syn-
channel interference estimation can be performed too, to chronization, the designed MATLAB code has some im-
assess the coexistence with nearby DVB-T (or LTE) chan- proved features not present in [19], such as LLR-based
nels [33]. In addition, power allocation algorithms can be soft-output demapping and FFT-based channel estimation.
exploited without changing the DVB-T receiver. Another With respect to the MATLAB DVB-T modulator of [23],
possibility is the investigation of the effect of impulsive there are two main differences. First, [23] does not include
noise [34], non-Gaussian channels, and quantization errors frame adaptation, and hence pilots and TPS are not in-
[35], on the DVB-T received signal. Furthermore, by cluded. Second, in [23], OFDM is not implemented.
means of our SDR-based design, the DVB-T signal can
first be received by a custom DVB-T receiver and then
retransmitted to a single destination by a regenerative re- 6. Conclusion
lay, nearby to the receiver. Coverage issues may be inves-
tigated by incorporating the designed software into a We have proposed two software implementations of
transmitting device for trial experiments. The proposed DVB-T, using C++ and MATLAB, publicly available
software could be used also jointly with cognitive radio under the GNU license. The generated DVB-T signal has
systems that try to exploit the TV white spaces: for in- been validated by the equivalence between C++ and
stance, in [36], the authors introduce a spectrum-sensing- MATLAB outputs and by the correct reception by TV sets
based system where HDTV signals received via a satellite and USB dongles. When incorporated into the Iris SDR
link can be redistributed in home environments over unoc- framework, the C++ DVB-T modulator works in real time.
cupied DVB-T channels, using a DVB-T transmitter. Ob- The proposed code can be useful for many applications and
viously, this mentioned list of possible applications is only can be easily modified. Future work could include DVB-
exemplificative and non-exhaustive. T2 [37], DVB-T2-Lite [38], and DVB-C2 [39].

5.2 Comparison with Other DVB-T Software

Herein we detail the main differences between the Acknowledgements
proposed DVB-T software and the existing literature. We
focus on those DVB-T designs that are closely related to The research leading to these results has received
our proposed design: about C++, we consider the solutions funding from the European Horizon 2020 Programme
proposed in [9] and [19]; with reference to MATLAB de- under grant agreement no. 645274 (WiSHFUL project,
signs, we compare with [19] and [23]. Open Call 1, DVB-TX-IRIS Extension). The authors thank
Nicholas Kaminski, Maicon Kist, Alextian Liberato, Diar-
With respect to the Soft-DVB in [9], there are four
muid Collins, Pedro Alvarez, and Francisco Paisana of
main differences. First, our C++ code is publicly available
TCD for all the help in using the Iris software and testbed.
(at https://github.com/wishful-project/module_iris/tree/
The authors also thank Gianluca Reali and Mauro
master/dvb-tx-iris/components/gpp/phy). Second, [9] is
Femminella for the USRP hardware. We also thank the
based on GNU Radio, while our C++ code is based on the
anonymous reviewers for their constructive comments.
Iris framework, which requires less memory. Third, [9]
employs a hybrid Python/C++ language, while our soft-
ware uses the C++ language only. Fourth, [9] produces RF
signals with 7 MHz channel bandwidth only (due to References
throughput limitations of the USB-based interface towards
the USRP in [9]), while our software can employ different [1] Digital Video BROADCASTING (DVB). Framing structure,
channel bandwidths, including the 8 MHz one. With re- channel coding and modulation for digital terrestrial television,
RADIOENGINEERING, VOL. 27, NO. 2, JUNE 2018 9

ETSI EN 300 744 V1.6.1 (2009-01). ETSI. In Proceedings of IEEE International Conference on Advanced
Technologies for Communication (ATC). Hanoi (Vietnam), 2012,
[2] POGGIONI, M., RUGINI, L., BANELLI, P. DVB-T/H and
5 p. DOI: 10.1109/ATC.2012.6404278
T-DMB: Physical layer performance comparison in fast mobile
channels. IEEE Transactions on Broadcasting, 2009, vol. 55. [18] LEE, K.-H., HEO, S. W. GPU based software DVB-T receiver
no. 4, p. 719–730. DOI: 10.1109/TBC.2009.2034418 design. In Proceedings of IEEE International Conference on
Consumer Electronics (ICCE). Las Vegas (USA), 2013, 5 p. DOI:
[3] REIMERS, U. H. DVB - The family of international standards for
10.1109/ICCE.2013.6487027
digital video broadcasting. Proceedings of the IEEE, 2006, vol. 94,
no. 1, p. 173–182. DOI: 10.1109/JPROC.2005.861004 [19] BARUFFA, G., RUGINI, L., BANELLI, P. Design and validation
of a software defined radio testbed for DVB-T transmission.
[4] ULVERSØY, T. Software defined radio: Challenges and opportu-
Radioengineering, 2014, vol. 23, no. 1, p. 387–398.
nities. IEEE Communications Surveys and Tutorials, 2010, vol. 12,
no. 4, p. 531–550. DOI: 10.1109/SURV.2010.032910.00019 [20] MACIEL, Y. P., AKAMINE, C., BEDICKS, JR. G., LOPES, P. B.
ISDB-TB transmission in software-defined radio. In Proceedings of
[5] PALKOVIC, M., RAGHAYAN, P., LI, M., et al. Future software-
the 7th IEEE Latin-American Conference on Communications
defined radio platforms and mapping flows. IEEE Signal
(LATINCOM). Arequipa, (Peru), 2015, 6 p. DOI:
Processing Magazine, 2010, vol. 27, no. 2, p. 22–33. DOI:
10.1109/LATINCOM.2015.7430122
10.1109/MSP.2009.935386
[21] CICHON, G., FETTWEIS, G., MOUSE: A shortcut from Matlab
[6] BLAKE, G., DRESLINSKI, R. G., MUDGE, T. A survey of
source to SIMD DSP assembly code. In Proceedings of
multicore processors. IEEE Signal Processing Magazine, 2009,
International Workshop on Embedded Computer Systems. Samos
vol. 26, no. 6, p. 26–37. DOI: 10.1109/MSP.2009.934110
(Greece), 2004, p. 159–167. DOI: 10.1007/978-3-540-27776-7_17
[7] KRATOCHVÍL, T., SLANINA, M. The DVB channel coding
[22] VEJRAŽKA, F., KOVAR, P., ESKA, M., PURICER, P. Software
application using the DSP development board MDS TM-13 IREF.
navigation receivers for GNSS and DVB. TransNav International
Radioengineering, 2004, vol. 13, no. 4, p. 14–17.
Journal on Marine Navigation and Safety of Sea Transportation,
[8] IANCU, D., YE, H., GLOSSNER, J., et al. Software-only 2007, vol. 1, no. 2, p. 137–141.
implementation of DBV-H. In Proceedings of SPIE Multimedia on
[23] HÜTTL, A., KRATOCHVÍL, T. DVB-T channel coding
Mobile Devices, 2008, 8 p. DOI: 10.1117/12.763800
implementation in MATLAB. In Proceedings of MATLAB
[9] PELLEGRINI, V., BACCI, G., LUISE, M. Soft-DVB: A fully- Conference. Prague (Czech Republic), 2009.
software GNU Radio-based ETSI DVB-T modulator. In
[24] POLÁK, L., KRATOCHVÍL, T. Simulation of the DVB-H
Proceedings of the 5th Karlsruhe Workshop on Software Radios,
channel coding and transmission in MATLAB. In Proceedings of
2008.
IEEE International Conference Radioelektronika. Brno (Czech
[10] PELLEGRINI, V., LUISE, M. Fully software OFDM modulation Republic), 2010, 4 p. DOI: 10.1109/RADIOELEK.2010.5478591
in vehicular, highly time-variant channels: An implemented
[25] SUTTON, P., LOTZE, J., LAHLOU, H., et al. Iris: an architecture
technology and its results. In Proceedings of IEEE International
for cognitive radio networking testbeds. IEEE Communications
Symposium on Wireless Communication Systems. Tuscany (Italy),
Magazine, 2010, vol. 48, no. 9, p. 114–122. DOI:
2009. DOI: 10.1109/ISWCS.2009.5285237
10.1109/MCOM.2010.5560595
[11] PELLEGRINI, V., DI DIO, M., ROSE, L., LUISE, M. On the
[26] DOYLE, L. E., SUTTON, P. D., NOLAN, K E., et al. Experiences
computation/memory trade-off in software defined radios. In
from the Iris testbed in dynamic spectrum access and cognitive
Proceedings of IEEE Global Telecommunication Conference
radio experimentation. In Proceedings of IEEE Symposium on New
(GLOBECOM). Miami (FL, USA), 2010. DOI:
Frontiers in Dynamic Spectrum (DySPAN). Singapore, 2010, 8 p.
10.1109/GLOCOM.2010.5683494
DOI: 10.1109/DYSPAN.2010.5457835
[12] JIANG, Y., XU, W., GRASSMANN, C. Implementing a DVB-T/H
[27] FRIGO, M., JOHNSON, S. G. FFTW: An adaptive software
receiver on a software-defined radio platform. International
architecture for the FFT. In Proceedings of IEEE International
Journal of Digital Multimedia Broadcasting, 2009, 7 p. DOI:
Conference on Acoustics, Speech and Signal Processing (ICASSP).
10.1155/2009/937848
Seattle (USA), 1998, p. 1381–1384. DOI:
[13] SUGANO, H., MIYAMOTO, R., OKADA, M. Fully software- 10.1109/ICASSP.1998.681704
based real-time ISDB-T 1 segment receiver. In Proceedings of [28] MARTIN, K., HOFFMAN, B. An open source approach to
IEEE International Symposium on Broadband Multimedia Systems developing software in a small organization. IEEE Software, 2007,
and Broadcasting (BMSB). Nuremberg (Germany), 2011, 5 p. vol. 24, no. 1, p. 46–53. DOI: 10.1109/MS.2007.5
DOI: 10.1109/BMSB.2011.5954956
[29] TOSATO, F., BISAGLIA, P. Simplified soft-output demapper for
[14] YU, J.-C., SHIH, J.-Z., HSU, Y.-T., TSENG, S.-M. Reed-Solomon binary interleaved COFDM with application to HIPERLAN/2. In
decoder optimization for PC-based DVB-T software radio Proceedings of IEEE International Conference on Communication
receiver. In Proceedings of IEEE International Conference on (ICC). New York (USA), 2002. DOI: 10.1109/ICC.2002.996940
Consumer Electronics (ICCE). Las Vegas (USA), 2011, 2 p.
DOI: 10.1109/ICCE.2011.5722645 [30] BARUFFA, G., RUGINI, L. Soft-output demapper with
approximated LLR for DVB-T2 systems. In Proc. IEEE Global
[15] TSENG, S.-M., HSU, Y.-T., LIN, H.-K. Iterative channel decoding Communication Conference (GLOBECOM). San Diego (USA),
for PC-based software radio DVB-T receiver. Wireless Personal 2015. DOI: 10.1109/GLOCOM.2015.7417395
Communications, 2013, vol. 69, no. 1, p. 403–411.
DOI: 10.1007/s11277-012-0580-z [31] GOMEZ-BARQUERO, D., WINSTON CALDWELL, M.
Broadcast television spectrum incentive auctions in the U.S.:
[16] TSENG, S.-M., HSU, Y.-T., CHANG, Y.-Y., LEE, T.-C. Software Trends, challenges, and opportunities. IEEE Communications
baseband optimization except channel decoding for PC-based Magazine, 2015, vol. 53, no. 7, p. 50–56. DOI:
DVB-T software radio receiver. Lecture Notes in Electrical 10.1109/MCOM.2015.7158265
Engineering, 2014, vol. 260, p. 319–327. DOI: 10.1007/978-94-
007-7262-5_37 [32] ALA-FOSSI, M., BONET, M. Clearing the skies: European
spectrum policy and future challenges of DTT in Finland and
[17] TSENG, S.-M., CHANG, T.-K., HSU, Y.-T. A/D USB dongle Spain. International Journal of Digital Television, 2016, vol. 7,
implementation for NB/PC-based software radio DVB-T receiver. no. 3, p. 363–377. DOI: 10.1386/jdtv.7.3.363_1
10 G. BARUFFA, L. RUGINI, F. FRESCURA, ET AL., REAL-TIME GENERATION OF STANDARD-COMPLIANT DVB-T SIGNALS

[33] BARUFFA, G., FEMMINELLA, M., MARIANI, F., REALI, G. signal processing for communications, with emphasis on
Protection ratio and antenna separation for DVB-T/LTE
multicarrier and cognitive radio techniques. He serves as
coexistence issues. IEEE Communications Letters, 2013, vol. 17,
no. 8, p. 1588–1591. DOI: 10.1109/LCOMM.2013.070113.130887 an Associate Editor of the IEEE Signal Processing Letters,
Digital Signal Processing, and the EURASIP Journal on
[34] BANELLI, P. Bayesian estimation of a Gaussian source in
Middleton’s class-A impulsive noise. IEEE Signal Processing
Advances in Signal Processing.
Letters, 2013, vol. 20, no. 10, p. 956–959. DOI:
Fabrizio FRESCURA is an Assistant Professor at the
10.1109/LSP.2013.2274774
Department of Engineering of the University of Perugia
[35] RUGINI, L., BANELLI, P. On the equivalence of maximum SNR since January 2005, where he is the lecturer of digital sig-
and MMSE estimation: applications to additive non-Gaussian
channels and quantized observations. IEEE Transactions on Signal
nal processing courses. He received the Laurea degree
Processing, 2016, vol. 64, no. 23, p. 6190–6199. (cum laude) in Electronic Engineering from the University
DOI: 10.1109/TSP.2016.2607152 of Perugia in 1992, and the Ph.D. degree in Telecommuni-
[36] FADDA, M., MURRONI, M., POPESCU, V. An unlicensed cations in 1997. He was the advising professor of the win-
indoor HDTV multi-vision system in the DTT bands. IEEE ning team of the Texas Instruments 1997 DSP Solution
Transactions on Broadcasting, 2012, vol. 58, no. 3, p. 338–346. Challenge. From 1999 to 2000 he has been working in
DOI: 10.1109/TBC.2012.2201559 Texas Instruments France. In 2000, he was co-founder of
[37] GRÖNROOS, S., NYBOM, K., BJÖRKVIST, J. Complexity the university spin-off company Digilab2000. Since 2001
analysis of software defined DVB-T2 physical layer. Analog he coordinates the research activity of DSPLAB of Univer-
Integrated Circuits and Signal Processing, 2011, vol. 69, no. 2-3,
sity of Perugia. The main activities are related to DSP for
p. 131–142. DOI: 10.1007/s10470-011-9724-4
Image/Video processing and transmission, and DSP im-
[38] SAMO, D. A., SLIMANI, M., BARUFFA, G., RUGINI, L. A per- plementation. In 2004, he was co-founder of the university
formance study of DVB-T2 and DVB-T2-Lite for mobile recep-
tion. Digital Signal Processing, 2015, vol. 37, p. 35–42. DOI:
spin-off company Mediatech. Since 2002 he joined the
10.1016/j.dsp.2014.11.002 standardization committee ISO/IEC JTC 1/SC 29/WG 1.
He has been co-editor and co-chair of the JPWL (JPEG
[39] HASSE, P., ROBERT, J. A software-based real-time DVB-C2
receiver. In Proceedings of IEEE International Symposium on 2000 Wireless) AHG (Ad Hoc Group) and co-editor of D-
Broadband Multimedia Systems and Broadcasting (BMSB). Cinema AHG. The scientific activity is proven by about 80
Nuremberg (Germany), 2011. DOI: 10.1109/BMSB.2011.5954935 papers published on peer-reviewed international journals,
proceedings of international conferences, and contributions
to international organization standards.
About the Authors ... Paolo BANELLI received the Laurea degree (cum laude)
in Electronics Engineering and the Ph.D. degree in Tele-
Giuseppe BARUFFA (corresponding author) was born in
communications from the University of Perugia, Italy, in
Perugia, Italy, in 1970. He received the Laurea degree in
1993 and 1998, respectively. In 2005, he was appointed as
Electronic Engineering and the Ph.D. degree in Telecom-
Associate Professor in the Department of Electronic and
munications from the University of Perugia in 1996 and
Information Engineering, University of Perugia, where he
2001, respectively. In 2005, he visited Swiss Federal Pol-
has been an Assistant Professor since 1998. In 2001, he
ytechnic of Lausanne (EPFL), Switzerland. Since 2005, he
joined the SpinComm group, as a Visiting Researcher, in
has been an Assistant Professor of Telecommunications
the Electrical and Computer Engineering Department,
with the Department of Engineering at the University of
University of Minnesota, Minneapolis. His research inter-
Perugia. He is the author of about 60 scientific papers. His
ests include signal processing for wireless communica-
main research interests include digital television broad-
tions, with emphasis on multicarrier transmissions, signal
casting techniques, joint source/channel coding for video,
processing for biomedical applications, spectrum sensing
analysis of intermodulations in nonlinear systems, and
for cognitive radio, waveform design for 5G communica-
GNSS /Cellular hybrid positioning techniques.
tions, and graph signal processing. He was a Member
Luca RUGINI was born in Perugia, Italy, in 1975. He (2011–2013) of the IEEE Signal Processing Society’s
received the Laurea degree (cum laude) in Electronic Engi- Signal Processing for Communications and Networking
neering in 2000 and the Ph. D. degree in Telecommunica- Technical Committee. In 2009, he was a General Co-chair
tions in 2003 from the University of Perugia, Perugia, of the IEEE International Symposium on Signal Processing
Italy. In 2007, he visited Delft University of Technology, Advances for Wireless Communications. He currently
Delft, The Netherlands. He is currently an Assistant Pro- serves as an Associate Editor of the IEEE Transactions on
fessor with the Department of Engineering at the Univer- Signal Processing and the EURASIP Journal on Advances
sity of Perugia. His research interests lie in the area of in Signal Processing.