Prof. Marco Giordani (marco.giordani@unipd.it)
• Marco Giordani (marco.giordani@unipd.it)
• Professore Associato @ DEI UNIPD.
• L. Triennale in Ingegneria dell’Informazione.
• L. Magistrale in Ingegneria delle Telecomunicazioni.
• Dottorato in Ingegneria delle Telecomunicazioni
• Reti wireless di nuova generazione (5G/6G).
• 32 Dipartimenti
• 70’000+ studenti
• 2700+ docenti
• 2700+ PTA
Docenti e Ricercatori
Studenti triennali
Dottorandi Budget di ricerca
Why your mobile-generation children have a totally different sign for phone: https://www.dailymail.co.uk/femail/article -8503415/Dad-shows-new-hand-gesturekids-use-theyre-phone.html
https://abcnews.go.com/International/leo-xiv-conclave-numbersnumber-cardinals-length-voting/story?id=121606316
Verso le nuove frontier delle tecnologie dell’informazione:
Prof. Marco Giordani (marco.giordani@unipd.it)
L’ORDINE INCONTRA I GIOVANI: Aperitivo Ingegneristico
20 giugno 2025 // Mestre (VE)
http://agcom.it
https://radar.cloudflare.com/traffic/ir?dateRange=7d
M. Giordani and M. Zorzi, "Non-Terrestrial Networks in the 6G Era: Challenges and Opportunities," in IEEE Network, vol. 35, no. 2, pp. 244-251, Mar. 2021.
“Volare basso”
Elevata flessibilità
Possono essere dispiegati “on demand”
Consumo elevato di potenza
Cono di copertura limitato
GEO
MEO/LEO
36’000 km di altezza
Raggio di copertura ENORME
Ritardo di propagazione
300÷15’000 km di altezza
Raggio di copertura GRANDE
Saturazione della capacità di cella
Non stazionario (deve operare in costellazioni)
Starlink IRIS2
Sistema privato
LEO (550 km)
In servizio (40€/mese)
~7000 satelliti (~42’000 in previsione)
Sistema UE (pubblica sicurezza)
LEO (2’000 km) + MEO (10’000 km)
In servizio dal 2035
290 satelliti (in previsione)
2.6 miliardi di USD 10 miliardi di Euro
Copertura globale
Copertura mirata per regioni strategiche UE
SatelliteUESatelliteUESatelliteUESatelliteUE
Carrierfrequency2GHz20GHz2GHz20GHz Bandwidth30MHz400MHz30MHz400MHz
DLFSPL159.1dB179.1dB190.6dB210.6dB
DLSNR6.6dB8.5dB0dB11.6dB
Altitude600kmN/A600kmN/A35786kmN/A35786kmN/A EIRP34dBW/MHz23dBm16dBW/MHz33dBm59dBW/MHz23dBm40dBW/MHz33dBm
Antennadiameter2mN/A0.5m0.6m22mN/A5m0.6m
Antennagain30dBi0dBi38.5dBi39.7dBi51dBi0dBi58.5dBi39.7dBi
Noisefigure-7dB-1.2dB-7dB-1.2dB Kaband(LEO) Kaband(GEO) Sband(LEO) Sband(GEO)
Fig.2: End-to-endthroughput,PDR,andlatencyattheapplicationlayervs.thesourcerate R .Wefocusonaregenerativepayloadarchitecture, andconsideraLEOsatelliteat h =600 kmvs.aGEOsatelliteat h =35786 km,atbothSandKabands.
Figure 3.3 presentstheaverageRTTmeasuredinbothterrestrialandnon-terrestrialnetworks, categorizedbydestinationandweatherconditions.ThebarsindicatesthemeanRTT,whilethe verticallinesindicatethevariance(i.e. µ ± ).
ItisevidentthattheterrestrialnetworkachievessignificantlylowerRTTsfor close destinations comparedtothenon-terrestrialnetwork.However,for far destinations,thenon-terrestrialnetwork performscomparablyorevenbetterthantheterrestrialnetwork.Thisoutcomeisexpected,asfor close destinations,thenon-terrestrialnetworkisdisadvantagedbytheinitialsatellitelink(extended bent-pipeapproach,seelaterinsubsection 3.4),whereasfor far destinations,leveragingsatellite connectivitybecomesbeneficial.It’simportanttonoticethatthevarianceinRTTforthenonterrestrialnetworkishigherthanthatoftheterrestrialnetwork,regardlessofdestination.
Figure9:Distributionofmedian(a)downloadand(b)upload goodputoverStarlinkfromselectedcitiesglobally.
• RTT: rete terrestre migliore per distanze brevi, ma Starlink è competitiva (o migliore) su lunghe distanze
Interestingly,theRTTvaluesinthenon-terrestrialnetworkappeartobeindependentofweather conditions. Figure3.3:PingaverageRTTbyWeatherandDestination
Table 1: Luna gaming results over 150 mins playtime. Values denote me dian±SD and the worst performer is highlighte d.
Figure10:UplinkZoomtra coveraterrestrial(left)and Starlink(right).Verticallinesshow15srecon gurations. achieve ⇡ 50–100Mbpsdownloadand ⇡ 4–12Mbpsuploadrates atthe75th percentile.Wealsodonot ndanycorrelationbetween baselinelatencies(seeFigure6)andupload/downloadgoodput, evidentfromthecontrastingcasesofDublinandManila.However, weobserveaninversecorrelationbetweenlossratesandgoodputs;increasingfrom4–8%atthe75th -percentile(seeFigure22in AppendixC).Seattle,notableforitslatencyperformance,records averagegoodputs.Consideringhighmeasurementdensityfrom thisregion,thetrendmightbeduetoStarlink’sinternalthrottling orload-balancingtopreventcongestion[68].Wealso ndthat
Figure 11: Cloud gaming over 5G ( left) and Starlink (right).
a: Bartolomeo Morellato Mohan, Nitinder, et al. "A multifaceted look at starlink performance." Proceedings of the ACM Web Conference 2024.
Vertical dashed lines show Starlink recon guration inter vals. which we trace to FEC (For ward Error Corre ction) packets that are fre quently sent in addition to raw vide o data (on average
±
Verso le nuove frontier delle tecnologie dell’informazione:
Prof. Marco Giordani (marco.giordani@unipd.it)
L’ORDINE INCONTRA I GIOVANI: Aperitivo Ingegneristico
20 giugno 2025 // Mestre (VE)
1’300’000 90%
ParameterD0/S0D0/S1D0/S2D11/S0D14/S0D14/S1D14/S2
Avg.size[MB]3.2041.5110.2350.0710.2020.1000.016
Encodingtime[ms]00023.328.212.971.95
Decodingtime[ms]00010.4813.575.810.72
RangeNet++inferencetime[ms]05656005656
Sourcerate[Mbps]256.3120.918.85.716.281.3
Throughput(d< 50 m)[Mbps]259.4124.320.25.617.38.61.4 Throughput(
Velocità media upload 4G LTE: ~10 Mbps
Velocità media upload 5G: ~20 Mbps.
https://www.speedtest.net/global-index/italy
Fig.6:Meanthroughputandconfidenceintervals(shadedareas)vs. d,andfordifferentHSCcompressionconfigurations.Solid(dashed)linesarerelativeto theuseofrealdatafromSemanticKITTI(statisticaltrafficmodels).
Octree(LOW) Octree(HIGH) G-PCC
PNG(Cartesian) J-LS(Cartesian)
• Compressione è causa di data processing.
L’impatto della compressione non è trascurabile.
Octree(LOW) Octree(HIGH) G-PCC
2DImage Cartesian Octree
2DVideo Spherical G-PCC
PNG(Cartesian) J-LS(Cartesian) MJ2(Spherical) PNG(Spherical) J-LS(Spherical) LZW(Spherical)
Figure3:PSNRfordifferent2Dvs.3Dcompressionmethods.“Image” Cartesianfilesencodethreegeometriccoordinatesasatrichannelimage,thususingabout1/3moreBPPsthaninthe sphericalmethods.Wealsotriedtoconvertthepointcloudinto sphericalcoordinatesusingtheradiusonly,thusrepresenting theLiDAR’sinputasasingle-channelimage.Whilethis approachpermitstoreducetheBPPsby2/3comparedto Cartesianfiles,thefinalcompressionratewasunsatisfactory.
Figure3:PSNRfordifferent2Dvs.3Dcompressionmethods.“Image” compressionisobtainedbyaveragingPNGandJ-LScompression. Cartesianfilesencodethreegeometriccoordinatesasatrichannelimage,thususingabout1/3moreBPPsthaninthe sphericalmethods.Wealsotriedtoconvertthepointcloudinto sphericalcoordinatesusingtheradiusonly,thusrepresenting theLiDAR’sinputasasingle-channelimage.Whilethis approachpermitstoreducetheBPPsby2/3comparedto Cartesianfiles,thefinalcompressionratewasunsatisfactory.
Octree(LOW) Octree(HIGH) G-PCC
PNG(Cartesian) J-LS(Cartesian) MJ2(Spherical)
PNG(Spherical) J-LS(Spherical) LZW(Spherical)
Figure4:Compression(above)anddecompression(below)timesfordifferent 2Dvs.3Dcompressionmethods. (De)compressiontime. TimelycompressionanddecomG-PCC
Third,Fig.2illustratesthatvideo-basedmethodslikeLZW
Third,Fig.2illustratesthatvideo-basedmethodslikeLZW
Figure4:Compression(above)anddecompression(below)timesfordifferent 2Dvs.3Dcompressionmethods. (De)compressiontime. Timelycompressionanddecom-
Figure3:PSNRfordifferent2Dvs.3Dcompressionmethods.“Image” compressionisobtainedbyaveragingPNGandJ-LScompression.
Epidermal lesions
example images from two disease classes. These test images highlight the difficulty of malignant versus benign discernment for the three medically
Epidermal lesions
Melanocytic lesions Melanocytic lesions (dermoscopy)
1.28 MILIONI
example images from two disease classes. These test images highlight the difficulty of malignant versus benign discernment for the three medically
“l’oggetto” è piccolo.
algoritmi di detection algoritmi di detection
Fig.5.Totalcompressionanddecompressiontimevs.thecompression )forDraco.
Fig.7.AP@0.70(%)forthecarclassvs.thecompressionconfiguration,for differentcodecsanddetectors.Weusethenotation“q xx”toindicatethatthe performanceofDracodependsonlyon q .
Fig.8.AP@0.50(%)forthepedestrianclassvs.thecompressionconfiguration,fordifferentcodecsanddetectors.Weusethenotation“q xx”toindicate thattheperformanceofDracodependsonlyon q .
Fig.9.Totalcompressionanddecompressiontimevs.AP@0.70(%)forthe carclassandthecompressedfilesize,fordifferentdetectors.Thedashedred linecorrespondstotheTDdelayrequirement,setto100msbasedonTableI. minore compressione minore compressione
F. Bragato, M. Neri, P. Testolina, M. Giordani, F. Battisti, “Teleoperated Driving: a New Challenge for 3D Object Detection in Compressed Point Clouds,” submitted to the IEEE Transactions on Intelligent Transportation Systems (T-ITS), 2025.
Preference for inaction
Sparing pedestrians
https://www.moralmachine.net
Male
Female
Female
Executive
Male
Sparing females
Preference
Sparing the ft
Male doctor
Female doctor
Female athlete
Executive
Sparing higher status
Executive
Sparing the lawful
Male
Sparing the young
Female
Sparing more characters
Sparing humans
Prof. Marco Giordani (marco.giordani@unipd.it)