Nunin Farko na 512-Launi Shift Keying Siginar Demodulation Ta Amfani da Neural Equalization don Sadarwar Kamara ta Optical

Teburin Abubuwan Ciki

1. Gabatarwa & Bayyani

Wannan takarda ta gabatar da gwajin gwaji mai ban mamaki na 512-Launi Shift Keying (512-CSK) don Sadarwar Kamara ta Optical (OCC). Babban nasarar ita ce demodulation na farko mara kuskure na irin wannan tsarin maɗaukaki mai girma a kan nisan mita 4, tare da shawo kan babban ƙalubalen crosstalk mara layi da ke cikin masu karɓa na kamara ta hanyar amfani da sabon abu na na'urar daidaita ta tushen hanyar sadarwa ta neural (NN) mai yawan lakabi.

OCC ana sanya ta a matsayin fasahar wayar tarho ta optical ta zamani, tana amfani da kayan aikin hoto na CMOS da ke ko'ina a cikin wayoyin hannu da na'urori. Wani babban bincike shine haɓaka ƙimar bayanai, wanda aka takura ta hanyar ƙimar firam ɗin kamara. CSK tana daidaita bayanai akan bambancin launi daga mai watsa RGB-LED, wanda aka tsara a cikin sararin launi na CIE 1931. CSK mai girma (misali, 512-CSK) yana yiwa alƙawarin ingantaccen ingantaccen yanayi amma yana fama da crosstalk tsakanin launuka saboda hankalin kamara da matatun launi.

2. Tsarin Fasaha

2.1 Saita Mai Karɓa & Kayan Aiki

Tsarin mai karɓa an gina shi a kusa da na'urar hoto ta Sony IMX530 CMOS, wanda aka zaɓa saboda ikonsa na fitar da bayanan RGB na danye na 12-bit ba tare da sarrafa bayanai ba (demosaicing, cire amo, daidaita farin). Wannan bayanan danye yana da mahimmanci don dawo da siginar daidai. Ana kama siginar ta hanyar ruwan tabarau na optical na 50mm. Mai watsa shi shine jeri na RGB-LED mai girma 8×8 (girman panel: 6.5 cm).

2.2 Sarrafa Siginar & Daidaita ta Neural

Hanyar sarrafawa ita ce kamar haka:

1. Samun Bayanan Danye: Kama ƙimar RGB da ba a sarrafa su ba daga na'urar.
2. Canza Sararin Launi: Canza RGB zuwa madaidaicin launi na CIE 1931 (x, y) ta amfani da matrix na yau da kullun: $\begin{pmatrix} x \\ y \end{pmatrix} = \begin{pmatrix} 0.4124 & 0.3576 & 0.1805 \\ 0.2126 & 0.7152 & 0.0722 \end{pmatrix} \begin{pmatrix} R \\ G \\ B \end{pmatrix}$.
3. Daidaita ta Hanyar Sadarwa ta Neural: Madaidaicin (x, y) ana shigar da su cikin NN mai yawan lakabi. An tsara wannan hanyar sadarwa don koyo da rama crosstalk mara layi tsakanin tashoshin launi. Tana da raka'a shigar 2 (x, y), $N_h$ ɓangarori na ɓoye tare da raka'a $N_u$, da raka'a M=9 na fitarwa (wanda ya dace da kowane alama na bit 9 don 512-CSK).
4. Demodulation & Decoding: Hanyar sadarwa ta NN tana fitar da rarraba yuwuwar bayan. Ana ƙididdige Ma'auni na Log-Likelihood (LLRs) daga wannan kuma a shigar da su cikin na'urar warware LDPC don gyaran kuskure na ƙarshe.

Alamomin 512-CSK an jera su a jere a cikin tsarin triangular a cikin zane na CIE 1931, farawa daga kusurwar shuɗi (x=0.1805, y=0.0722).

3. Sakamakon Gwaji & Bincike

3.1 Ayyukan BER vs. Girman Jerin LED

Gwajin ya bambanta adadin LED masu aiki a cikin jeri daga 1×1 zuwa 8×8 don kimanta ƙimar Kuskuren Bit (BER) a matsayin aikin ƙarfin hashen da aka karɓa (yanki a cikin hoto). An tsara nisan watsawa a mita 4. Sakamakon ya nuna cewa na'urar daidaita ta neural tana da mahimmanci don cimma aiki mara kuskure tare da cikakken jeri na 8×8, yana magance crosstalk da ke ƙaruwa tare da ƙarfin siginar da yanki.

3.2 Ma'auni Mafi Muhimmanci na Aiki

• Matsayin Maɗaukaki: 512-CSK (9 bits/symbol), rikodin mafi girma don nunin gwaji na OCC.
• Nisa: Mita 4, yana nuna kewayon aiki.
• Mai Ba da Damar: Daidaita mara layi na tushen hanyar sadarwa ta neural da aka yi amfani da shi kai tsaye ga bayanan danye na na'urar.
• Kwatanta: Wannan aikin ya ci gaba sosai fiye da nunin da suka gabata (8-CSK, 16-CSK, 32-CSK) a cikin duka matsayin maɗaukaki da ingantaccen fasahar ramawa.

4. Bincike na Tsaki & Fassarar Kwararru

Hankali na Tsaki: Wannan takarda ba kawai game da tura CSK zuwa launuka 512 ba ne; tabbataccen hujja ce cewa sarrafa siginar ta hanyar bayanai, ta hanyar sadarwa ta neural shine mabuɗin buɗe ingantaccen OCC. Marubutan sun gano daidai cewa babban toshewa ba LED ko na'urar ba ne, amma rikice-rikice, mara layi a cikin tashar. Maganinsu—tsallake na'urorin daidaita layi na al'ada don NN mai yawan lakabi—canji ne mai ƙarfi da ƙarfi a falsafar ƙira, yana kwatanta nasarar masu karɓa na neural a cikin sadarwar RF [1].

Kwararar Ma'ana: Ma'ana tana da ƙarfi: 1) Ana buƙatar CSK mai girma don sauri, 2) Crosstalk na kamara yana kashe CSK mai girma, 3) Wannan crosstalk yana da rikitarwa kuma mara layi, 4) Saboda haka, yi amfani da mai kiyasin aiki na duniya (hanyar sadarwa ta neural) don soke shi. Amfani da bayanan danye na na'urar wani mahimmin bayani ne, wanda sau da yawa ake yin watsi da shi. Yana guje wa asarar bayanai da kuma rikice-rikicen da aka gabatar na na'urar sarrafa siginar hoto (ISP) na kamara, wanda ya dace da mafi kyawun ayyuka a cikin binciken daukar hoto na lissafi daga cibiyoyi kamar MIT Media Lab.

Ƙarfi & Kurakurai: Babban ƙarfin shi ne nasarar haɗa wani ɓangaren ML na zamani a cikin tarin sadarwa na zahiri, yana cimma rikodin da aka bayyana. Gwajin gwaji yana da haske. Duk da haka, binciken yana da kurakurai kamar na nunin farko: Babu ambaton ƙimar bayanai (bits/sec), kawai ingantaccen yanayi (bits/symbol). Tasirin kwararar gaske ya kasance maras tabbas. Bugu da ƙari, rikitarwar NN, buƙatun bayanan horo, da ikon haɓakawa zuwa kamarori ko yanayi daban-daban ba a bincika su ba—manyan cikas ga daidaitawa da kasuwanci.

Hankali Mai Aiki: Ga masu bincike, hanyar tana da haske: Mayar da hankali kan tsarin gine-gine masu sauƙi, masu daidaitawa na neural don daidaita lokacin gaskiya. Benchmarking ya kamata ya haɗa da ainihin kwarara da jinkiri. Ga masana'antu (misali, IEEE P802.15.7r1 OCC Task Group), wannan aikin yana ba da shaida mai ƙarfi don yin la'akari da masu karɓa na tushen neural a cikin ma'auni na gaba, amma dole ne a haɗa su da gwaji mai ƙarfi na haɗin kai. Mataki na gaba shine matsar daga saitin dakin gwaji mai tsayi zuwa yanayi mai motsi, watakila ta amfani da fasahohin da aka yi wahayi ta hanyar daukar yanki irin na CycleGAN [2] don barin NN ya rama yanayin hasken muhalli daban-daban, ƙalubale mai wuyar gaske fiye da crosstalk mai tsayi.

5. Cikakkun Bayanai na Fasaha & Tsarin Lissafi

Sarrafa siginar na tsaki ya ƙunshi manyan canje-canje guda biyu:

1. Canzawa daga RGB zuwa CIE 1931: $\begin{pmatrix} x \\ y \end{pmatrix} = \mathbf{M} \cdot \begin{pmatrix} R \\ G \\ B \end{pmatrix}$ inda $\mathbf{M}$ shine matrix da aka riga aka tsara: $\mathbf{M} = \begin{pmatrix} 0.4124 & 0.3576 & 0.1805 \\ 0.2126 & 0.7152 & 0.0722 \end{pmatrix}$. Wannan yana tsara ƙimar RGB mai dogaro da na'urar zuwa sararin launi na cikakke.

2. Hanyar Sadarwa ta Neural a matsayin Mai Daidaitawa: NN tana koyon aikin $f_{\theta}$ wanda ke tsara madaidaicin da aka ɓata da aka karɓa $(x', y')$ zuwa yuwuwar bayan $P(\text{symbol}_i | x', y')$ ga duk alamomi 512. Ana horar da sigogin $\theta$ don rage asarar giciye tsakanin yuwuwar da aka annabta da alamomin da aka watsa da aka sani. Ana kuma ƙididdige LLR don bit na $k$ kamar haka: $LLR(b_k) \approx \log \frac{\sum_{i \in S_k^1} P(\text{symbol}_i | x', y')}{\sum_{i \in S_k^0} P(\text{symbol}_i | x', y')}$ inda $S_k^1$ da $S_k^0$ su ne tarin alamomi inda bit na $k$ ya kasance 1 da 0, bi da bi.

6. Tsarin Bincike & Misalin Lamari

Tsarin don Kimanta Ci gaban OCC: Don tantance kowace sabuwar takardar OCC, muna ba da shawarar tsarin bincike mai girma huɗu:

1. Ingantaccen Yanayi na Spatio-Spectral (Bits/Albarkatu): Menene ƙimar bayanai da aka cimma (bps) kuma wane albarkatu ne yake amfani da shi (bandwidth, pixels na sarari, lokaci)? Wannan takarda tana da maki mai girma akan ingantaccen yanayi (bits/symbol) amma ba ta da adadi na bps.
2. Ƙarfi & Aiki: Menene ƙuntatawa na aiki (nisa, daidaitawa, hasken muhalli)? 4m yana da kyau, amma yanayin tsaye yana da iyaka.
3. Rikitarwar Tsarin & Farashi: Menene farashin maganin? Na'urar daidaita ta neural tana ƙara farashin lissafi da horo.
4. Yiwuwar Daidaitawa: Yaya ake sake yin fasahar da kuma haɗin kai? Dogaro da bayanan danye da NN da aka horar a halin yanzu yana rage wannan maki.

Misalin Lamari - Aiwatar da Tsarin: Kwatanta wannan aikin 512-CSK NN tare da aikin 8-CSK na al'ada ta amfani da daidaita layi [3].

• Inganci: 512-CSK ya fi girma a cikin bits/symbol.
• Ƙarfi: NN na iya sarrafa mara layi da kyau, amma aikin sa a ƙarƙashin yanayin da ba a horar da shi ba (sabon kamara, haske daban) ba a sani ba idan aka kwatanta da mafi sauƙi na layi.
• Rikitarwa: NN yana da rikitarwa sosai.
• Daidaitawa: Daidaita layi yana da sauƙin daidaitawa.

Musayar tana da haske: ci gaban sarrafa siginar yana siyan ingantaccen inganci a farashin rikitarwa. Hanyar fagen ita ce karɓar wannan rikitarwa don shawo kan iyakokin zahiri.

7. Aikace-aikace na Gaba & Hanyoyin Bincike

Tasirin wannan aikin ya wuce dakin gwaji:

• LiFi Mai Girma-Girma don 6G: Haɗa irin wannan OCC mai girma tare da abubuwan more rayuwa na LiFi na iya samar da damar shiga hotspot na gigabit da yawa a cikin filayen wasa, filayen jiragen sama, ko masana'antu masu wayo, tare da haɗa hanyoyin sadarwa na RF.
• IoT Mai Tsakiya akan Wayar Hannu: Ba da damar musayar bayanai mai aminci, kusa da kusa (misali, biyan kuɗi, tikitin, haɗa na'urori) ta amfani da kamarorin wayar hannu a matsayin masu karɓa tare da ƙaramin ƙarin kayan aiki.
• Sadarwar Motoci V2X: Yin amfani da fitilun mota/fitilun baya da kamarori don sadarwa kai tsaye tsakanin mota-da-mota ko mota-da-abubuwan more rayuwa, haɓaka tsarin aminci.

Hanyoyin Bincike Masu Muhimmanci:

1. Koyo Mai Daidaitawa & Haɗin Kai don Masu Daidaitawa: Haɓaka NNs waɗanda zasu iya daidaita kan layi zuwa sabbin samfuran kamara ko haske, yuwuwar ta amfani da koyo a cikin tarin na'urori don gina samfura masu ƙarfi ba tare da raba bayanan danye ba.
2. Haɗin Coding na Tushe-Tashar tare da Hangen Nesa: Bincika fasahohin koyo mai zurfi waɗanda ke haɗa maɗaukaki (tsarin CSK) da na'urar daidaitawa don takamaiman na'urar kamara, kama da tsarin sadarwa da aka koya daga ƙarshe zuwa ƙarshe.
3. Inganta Haɗin Layer: Haɗa na'urar daidaita NN na Layer na zahiri tare da ƙa'idodin Layer mafi girma don inganta jimlar kwararar tsarin da aminci a cikin yanayi mai motsi.

Haɗuwar sadarwa, hangen nesa na kwamfuta, da koyon injin, kamar yadda aka nuna a cikin wannan takarda, shine inda za a fitar da mafi girman sabbin abubuwa a cikin OCC.

8. Nassoshi

1. O'Shea, T. J., & Hoydis, J. (2017). Gabatarwa ga Koyo Mai Zurfi don Layer na Zahiri. IEEE Transactions on Cognitive Communications and Networking. (Misalin hanyoyin sadarwa na neural a cikin sadarwa).
2. Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Fassarar Hotuna-zuwa-Hoto mara Haɗin gwiwa ta amfani da Hanyoyin Sadarwa na Adawa na Ci gaba da Ci gaba. Proceedings of the IEEE International Conference on Computer Vision (ICCV). (CycleGAN don ɗaukar yanki).
3. Chen, H.-W., et al. (2019). [1] a cikin ainihin PDF. (Misalin aikin CSK na farko, ƙananan matsayi).
4. IEEE Standard for Local and Metropolitan Area Networks--Part 15.7: Short-Range Optical Wireless Communications. IEEE Std 802.15.7-2018.
5. MIT Media Lab, Computational Photography. (Tushen ra'ayi don mahimmanci na bayanan danye na na'urar).